Signal modulation type identification method, identification model training method and system

Abstract

The embodiment of the invention provides a signal modulation type identification method and system and an identification model training method and system.The signal modulation type identification method comprises the steps that a target signal image is input into a trained identification model, and a target modulation type is obtained through identification of the identification model, the target modulation type is the modulation type of a target signal described by the target signal image. The recognition process comprises the steps of performing feature extraction on a target signal image through a first feature extraction network of a recognition model to obtain a first image feature, and recognizing a target modulation type based on the first image feature through a classification network, the first feature extraction network is a network which is pre-trained by using a first image set and has an image understanding capability, each image in the first image set is a non-signal image, the training process of the recognition model is based on a second image set, and each image in the second image set is a signal image.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and particularly to a method for identifying signal modulation types, a method for training an identification model, and a system. Background Art

[0002] Currently, there is a need to identify the modulation types of signals in many fields. Accurately identifying the modulation type of a signal is not only crucial for the performance of the system, but also directly affects the reliability and security of information transmission.

[0003] Traditional identification methods usually rely on features extracted from the signal itself, such as extracting the instantaneous frequency, phase, and amplitude of the signal, and identifying the modulation type of the signal based on the extracted signal features. However, the performance of these features varies greatly in different environments and is easily affected by factors such as noise and multipath effects. Therefore, the identification rate of the signal modulation type identification method in the prior art is relatively low, and its robustness and accuracy are limited to a certain extent.

[0004] The content in the background art section is only information known to the inventor personally, and does not represent that the above information has entered the public domain before the filing date of this disclosure, nor does it represent that it can become the prior art of this disclosure. Summary of the Invention

[0005] This specification provides a method for identifying signal modulation types, a method for training an identification model, and a system, which can improve the identification accuracy of signal modulation type identification and enhance the identification speed.

[0006] In a first aspect, this specification provides a method for identifying signal modulation types, including: obtaining a target signal image; inputting the target signal image into a trained identification model to identify a target modulation type through the identification model, where the target modulation type is the modulation type of the target signal described by the target signal image. The identification model at least includes a first feature extraction network and a classification network. The identification process of the identification model includes: extracting first image features from the target signal image through the first feature extraction network, and identifying the target modulation type based on the first image features through the classification network. Before the identification model is trained, the first feature extraction network is a network with image understanding ability pre-trained using a first image set, and each image in the first image set is a non-signal image. The training process of the identification model is based on a second image set, and each image in the second image set is a signal image.

[0007] In some embodiments, the recognition model further includes a second feature extraction network, and the recognition process of the recognition model further includes: extracting second image features from the target signal image through the second feature extraction network; recognizing the target modulation type based on the first image features through the classification network, including: recognizing the target modulation type based on the first image features and the second image features through the classification network.

[0008] In some embodiments, the first feature extraction network and the second feature extraction network satisfy at least one of the following: the number of network parameters of the first feature extraction network is greater than that of the second feature extraction network, the network depth of the first feature extraction network is greater than that of the second feature extraction network, or the network width of the first feature extraction network is greater than that of the second feature extraction network.

[0009] In some embodiments, the first image features are features obtained by understanding the content in the target signal image and encoding the understood content; the second image features are features obtained by encoding the target signal image from a high-dimensional space to a low-dimensional space.

[0010] In some embodiments, recognizing the target modulation type based on the first image features and the second image features through the classification network includes: performing alignment processing on the first image features and the second image features through the classification network to make the dimensions of the first image features and the second image features equal; performing fusion processing on the aligned first image features and the second image features through the classification network to obtain fusion features; and performing classification processing on the fusion features through the classification network to obtain the target modulation type.

[0011] In some embodiments, performing fusion processing on the aligned first image features and the second image features through the classification network to obtain fusion features includes: determining a first weight corresponding to the first image features and a second weight corresponding to the second image features; and performing weighted calculation on the first image features and the second image features according to the first weight and the second weight to obtain the fusion features.

[0012] In some embodiments, extracting first image features from the target signal image through the first feature extraction network includes: performing shallow feature extraction and deep feature extraction on the target signal image through the first feature extraction network to obtain shallow features and deep features corresponding to the target signal image; and performing multi-scale feature fusion on the shallow features and the deep features through the first feature extraction network to obtain the first image features.

[0013] In some embodiments, the classification network is any one of the following: an encoder-decoder network based on a self-attention mechanism; an encoder-decoder network with skip connections; or a multi-layer perceptron network.

[0014] In some embodiments, the target signal is a signal emitted by a radar, and the method further includes: determining the type of the radar based on the target modulation type.

[0015] In some embodiments, the target signal is a communication signal, and the method further includes: demodulating and analyzing the content of the target signal based on the target modulation type.

[0016] In a second aspect, the present specification also provides a method for training an identification model, including: obtaining an identification model to be trained, where the identification model at least includes a first feature extraction network and a classification network, the first feature extraction network is a network with image understanding ability pre-trained using a first image set, and each image in the first image set is a non-signal image; obtaining a second image set, where the second image set includes a plurality of sample signal images and annotation information corresponding to each sample signal image, and the annotation information represents the actual modulation type adopted by the sample signal described by the sample signal image; and performing multiple iterative trainings on the identification model using the second image set to obtain a trained identification model, where the process of each iterative training includes: extracting first image features from the sample signal image through the first feature extraction network, identifying a predicted modulation type corresponding to the sample signal through the classification network based on the first image features, and updating the parameters of the identification model with the training objective of minimizing the difference between the predicted modulation type and the actual modulation type.

[0017] In some embodiments, the identification model further includes a second feature extraction network, and the process of each iterative training further includes: extracting second image features from the sample signal image through the second feature extraction network; identifying a predicted modulation type corresponding to the sample signal through the classification network based on the first image features, including: identifying the predicted modulation type through the classification network based on the first image features and the second image features.

[0018] In some embodiments, the first feature extraction network and the second feature extraction network satisfy at least one of the following: the number of network parameters of the first feature extraction network is greater than the number of network parameters of the second feature extraction network, the network depth of the first feature extraction network is greater than the network depth of the second feature extraction network, or the network width of the first feature extraction network is greater than the network width of the second feature extraction network.

[0019] In some embodiments, the first image feature is a feature obtained by understanding the content in the sample signal image and encoding the understood content; the second image feature is a feature obtained by encoding the sample signal image from a high-dimensional space to a low-dimensional space.

[0020] In some embodiments, the classification network identifies the predicted modulation type based on the first image feature and the second image feature, including: aligning the first image feature and the second image feature by the classification network so that the dimensions of the first image feature and the second image feature are equal; fusing the aligned first image feature and the second image feature by the classification network to obtain a fused feature; and classifying the fused feature by the classification network to obtain the predicted modulation type.

[0021] In some embodiments, fusing the aligned first image feature and the second image feature by the classification network to obtain a fused feature includes: determining a first weight corresponding to the first image feature and a second weight corresponding to the second image feature; and performing weighted calculation on the first image feature and the second image feature according to the first weight and the second weight to obtain the fused feature.

[0022] In some embodiments, the first feature extraction network extracts a first image feature from the sample signal image, including: performing shallow feature extraction and deep feature extraction on the sample signal image by the first feature extraction network to obtain a shallow feature and a deep feature corresponding to the sample signal image; and performing multi-scale feature fusion on the shallow feature and the deep feature by the first feature extraction network to obtain the first image feature.

[0023] In some embodiments, taking minimizing the difference between the predicted modulation type and the actual modulation type as a training objective to update the parameters of the recognition model includes: determining the error between the predicted modulation type and the actual modulation type; determining the gradient of the error with respect to the parameters of the recognition model, and adjusting the parameters of the recognition model based on the gradient.

[0024] In a third aspect, this specification also provides a signal modulation type recognition system, including: at least one storage medium storing at least one instruction set for recognizing a signal modulation type; and at least one processor communicatively connected to the at least one storage medium. When the signal modulation type recognition system runs, the at least one processor reads the at least one instruction set and executes the signal modulation type recognition method according to any one of the first aspect as instructed by the at least one instruction set.

[0025] In a fourth aspect, this specification also provides a recognition model training system, including: at least one storage medium storing at least one instruction set for training a recognition model; and at least one processor communicatively connected to the at least one storage medium. When the recognition model training system runs, the at least one processor reads the at least one instruction set and executes the recognition model training method according to any one of the second aspect as instructed by the at least one instruction set.

[0026] As can be seen from the above technical solutions, for the signal modulation type recognition method, recognition model training method and system provided in this specification, the target signal is converted into the image domain to obtain the target signal image, and the target modulation type is obtained by recognizing the target signal image through the recognition model. Different from directly extracting features of the signal in the signal domain and recognizing the signal modulation type based on the extracted features in the prior art, in this specification, the differences of the signal are visualized through the image, and the features of the signal are automatically extracted from the target signal image, avoiding the limitations and errors of manual analysis and improving the signal recognition accuracy. And by converting the signal into an image form for analysis, the model can identify potential useful signal features in the noise, enhancing the robustness in complex environments. Further, the first feature extraction network in the recognition model is a network with image understanding ability pre-trained using non-signal images (i.e., the images in the first image set), which enables the recognition model to understand the target signal image and extract richer features based on the understood content, thereby improving the accuracy of the recognition result. On this basis, the recognition model is further trained more specifically based on the signal images converted from the sample signals (i.e., the images in the second image set), which can enhance the accuracy of the recognition model for signal image recognition while ensuring the generalization ability of the model and enhancing the overall performance ability of the recognition model.

[0027] Some of the other functions of the signal modulation type recognition method, recognition model training method and system provided in this specification will be listed below. The creative aspects of the signal modulation type recognition method, recognition model training method and system provided in this specification can be fully explained through practice or use of the methods, devices and combinations described in the detailed examples below. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] To more clearly illustrate the technical solutions in the embodiments of this specification, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of this specification. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0029] Figure 1 The figure shows a schematic diagram of an application scenario of a method for identifying a signal modulation type provided according to an embodiment of this specification;

[0030] Figure 2 The figure shows a hardware structure diagram of a method for identifying a signal modulation type provided according to an embodiment of this specification;

[0031] Figure 3 The figure shows a flowchart of a method for identifying a signal modulation type provided according to an embodiment of this specification;

[0032] Figure 4 The figure shows a schematic diagram of some images obtained by signal conversion provided according to an embodiment of this specification;

[0033] Figure 5 The figure shows an identification process of a signal modulation type provided according to an embodiment of this specification;

[0034] Figure 6 The figure shows another identification process of a signal modulation type provided according to an embodiment of this specification; and

[0035] Figure 7 The figure shows a flowchart of a method for training an identification model provided according to an embodiment of this specification. Detailed implementation manners

[0036] The following description provides specific application scenarios and requirements of this specification, aiming to enable those skilled in the art to manufacture and use the content in this specification. For those skilled in the art, various partial modifications to the disclosed embodiments are obvious, and the general principles defined here can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the shown embodiments, but has the widest scope consistent with the claims.

[0037] The terms used herein are for the purpose of describing particular example embodiments only and are not limiting. For example, unless the context clearly dictates otherwise, as used herein, the singular forms "a", "an" and "the" may also include the plural forms. When used in this specification, the terms "comprises", "comprising" and / or "having" mean that the associated integers, steps, operations, elements and / or components are present, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof in the system / method.

[0038] In view of the following description, these and other features of the present specification, as well as the operations and functions of the related elements of the structure, and the combination and manufacturing economy of the components can be significantly improved. Referring to the accompanying drawings, all of which form a part of this specification. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0039] The flowcharts used in this specification illustrate the operations implemented by the system according to some embodiments of this specification. It should be clearly understood that the operations of the flowchart may not be implemented in sequence. On the contrary, the operations may be implemented in reverse order or simultaneously. In addition, one or more other operations may be added to the flowchart. One or more operations may be removed from the flowchart.

[0040] For convenience of description, the terms that will appear hereinafter in this specification are first explained.

[0041] Wireless signal: refers to radio waves or other electromagnetic waves transmitted through a wireless medium (such as air, vacuum, etc.) and used to carry information signals. In this specification, a signal refers to a modulated wireless signal.

[0042] Modulation: refers to the process of embedding baseband information into a high-frequency carrier signal so that the information can be transmitted in a wireless channel.

[0043] Modulation type: It refers to the technology of modulating signals in different ways. According to different modulation technologies, the modulation type can be at least one of Binary Phase Shift Keying (BPSK), Circular Polarization (CP), Costas Loop, 4-Level Frequency Shift Keying with Baker 5 Algorithm (FSK4_Baker5), 4-Level Frequency Shift Keying with Linear Frequency Modulation (FSK4_LFM), Frank Modulation, Linear Frequency Modulation (LFM), Non-Linear Frequency Modulation (NLFM), polyphase codes (P1, P2, P3, P4), and multi-time codes (T1, T2, T3, T4). Each modulation type has different transmission characteristics and application scenarios.

[0044] The application scenarios of this specification are introduced below.

[0045] The embodiments of this specification can be applied to scenarios that require the identification of the modulation type of signals. For example, the embodiments of this specification can be used for detection. In this scenario, the method of the embodiments in this specification is used to identify the modulation type of intercepted radar signals. By identifying the modulation type of the radar signals, it is determined whether the intercepted radar signals are friendly signals, thereby helping to analyze the signal source. Another example is that the embodiments of this specification can be used for signal analysis in the communication field. In this scenario, after the receiving end receives the target signal, the method provided by the embodiments of this specification can be used to identify the target modulation type corresponding to the target signal. Then, the target signal is demodulated and analyzed according to the target modulation type, and the transmission parameters can be dynamically adjusted based on the modulation type to improve the communication quality with the sending end. In addition, by identifying the modulation type, abnormal or illegal signals in the communication signals can be detected to prevent potential threats.

[0046] It should be noted that the above application scenarios of signal recognition are only some examples among the multiple usage scenarios provided in this specification. The recognition of signal modulation types provided in this specification can be applied not only to the scenarios listed above, but also to all scenarios that require the recognition of signal modulation types. Those skilled in the art should understand that when the signal modulation type recognition method provided in this specification is applied to other usage scenarios, its implementation manner and technical effects are similar.

[0047] Figure 1 Fig. shows a schematic diagram of an application scenario for the recognition of a signal modulation type provided according to an embodiment of this specification. As Figure 1 shown, this application scenario 100 may include: a training system 110 for the recognition model (hereinafter referred to as the training system 110) and a recognition system 120 for the signal modulation type (hereinafter referred to as the recognition system 120).

[0048] Referring to Figure 1 , the application scenario 100 may involve two stages, namely the training stage and the inference stage.

[0049] In the training stage, the training system 110 may train the recognition model to be trained based on multiple sample signal images in the sample signal image set (i.e., the second image set) to obtain a trained recognition model. The trained recognition model has the ability to recognize the target modulation type. The training system 110 may deploy the trained recognition model in the recognition system 120.

[0050] In the inference stage, the recognition system 120 has a recognition model pre-deployed. When it is necessary to recognize the modulation type of the target signal, the recognition system 120 may obtain a target signal image for describing the target signal and input the target signal image into the recognition model to recognize the target modulation type through the recognition model.

[0051] In some embodiments, the training system 110 may store data and instructions for implementing the training method of the recognition model provided in this specification and may execute or be used to execute the data and instructions. In some embodiments, the training system 110 may include a hardware device with data information processing capabilities and necessary programs for driving the hardware device to work.

[0052] In some embodiments, the recognition system 120 may store data and instructions for implementing the signal modulation type recognition method provided in this specification and may execute or be used to execute the data and instructions. In some embodiments, the recognition system 120 may include a hardware device with data information processing capabilities and necessary programs for driving the hardware device to work.

[0053] It can be understood that the training system 110 and the recognition system 120 can correspond to the same system or different systems, and this specification does not limit this.

[0054] It should be noted that the training system 110 can correspond to a single device or a cluster of devices, and this specification does not limit this. When the training system 110 corresponds to a single device, the training method of the recognition model can be fully executed on this device. When the training system 110 corresponds to a cluster of devices, the training method of the recognition model can be executed cooperatively on multiple devices corresponding to the cluster of devices, and this specification does not limit this.

[0055] The recognition system 120 can correspond to a single device or a cluster of devices, and this specification does not limit this. When the recognition system 120 corresponds to a single device, the recognition method of the signal modulation type can be fully executed on this device. When the recognition system 120 corresponds to a cluster of devices, the recognition method of the signal modulation type can be executed cooperatively on multiple devices corresponding to the cluster of devices, and this specification does not limit this.

[0056] It should be noted that the user data obtained in this specification has all been authorized by the user and does not involve user privacy.

[0057] Figure 2 The hardware structure diagram of a computing system 200 provided according to an embodiment of this specification is shown. The computing system 200 can be used as Figure 1 the training system 110 in Figure 1 and execute the training method of the recognition model described in this specification. The computing system 200 can also be used as

[0058] As Figure 2 shown, the computing system 200 can include at least one storage medium 230 and at least one processor 220. In some embodiments, the computing system 200 can also include a communication port 250 and an internal communication bus 210. The computing system 200 can also include I / O components 260.

[0059] The internal communication bus 210 can connect different system components. For example, the internal communication bus 210 can connect the storage medium 230, the processor 220, the communication port 250, and the I / O components 260, etc.

[0060] The I / O components 260 support input / output between the computing system 200 and other components.

[0061] The communication port 250 is used for data communication between the computing system 200 and the outside world. For example, the communication port 250 can be used for data communication between the computing system 200 and the network 140. The communication port 250 can be a wired communication port or a wireless communication port.

[0062] The storage medium 230 can include a data storage device. The data storage device can be a non-transitory storage medium or a transitory storage medium. For example, the data storage device can include one or more of a magnetic disk 232, a read-only storage medium (ROM) 234, or a random access storage medium (RAM) 236. The storage medium 230 also includes at least one instruction set stored in the data storage device. The instruction set can include computer program code, and the computer program code can include programs, routines, objects, components, data structures, procedures, modules, and so on.

[0063] At least one processor 220 can be communicatively connected to at least one storage medium 230. When the computing system 200 runs, at least one processor 220 reads the at least one instruction set and, according to the instructions of the at least one instruction set, executes the method for identifying the signal modulation type provided in this specification, or executes the method for training the identification model provided in this specification. The processor 220 can execute the steps included in the above methods. The processor 220 can be in the form of one or more processors. In some embodiments, the processor 220 can include one or more hardware processors, such as a microcontroller, a microprocessor, a reduced instruction set computer (RISC), an application-specific integrated circuit (ASIC), an application-specific instruction set processor (ASIP), a central processing unit (CPU), a graphics processing unit (GPU), a physics processing unit (PPU), a microcontroller unit, a digital signal processor (DSP), a field-programmable gate array (FPGA), an advanced RISC machine (ARM), a programmable logic device (PLD), any circuit or processor capable of performing one or more functions, or any combination thereof.

[0064] For illustrative purposes only, only one processor 220 is shown for the computing system 200 in the drawings. However, it should be noted that the computing system 200 in this specification can also include multiple processors. Therefore, the operations and / or method steps disclosed in this specification can be executed by one processor or jointly executed by multiple processors. For example, if it is described in this specification that the processor 220 of the computing system 200 executes step A and step B, it should be understood that step A and step B can also be jointly or separately executed by two different processors 220 (for example, the first processor executes step A, the second processor executes step B, or the first and second processors jointly execute steps A and B).

[0065] Figure 3 FIG. 1 shows a flowchart of a method P300 for identifying a signal modulation type provided according to an embodiment of the present specification. As described above, the identification system 110 may execute the method P300 for identifying the signal modulation type of the present specification. Specifically, the processor 220 in the identification system 110 may read the instruction set stored in its local storage medium, and then execute the method P300 for identifying the signal modulation type of the present specification according to the provisions of the instruction set. As Figure 3 shown, the method P300 may include steps S310-S320:

[0066] S310: Obtain a target signal image.

[0067] In some embodiments, the target signal image may be an image used to reflect / describe the waveform of the target signal and can be obtained by converting the target signal. For example, it may be a time-domain waveform diagram, a frequency-domain waveform diagram, a time-frequency diagram, or a vector signal diagram, etc. The above different types of images can reflect the waveforms of different characteristics of the target signal. For example, the time-domain diagram shows the amplitude of the target signal changing with time; the frequency-domain diagram is obtained by converting the time-domain signal to the frequency-domain representation through Fourier transform and shows the power distribution of the signal at each frequency; the time-frequency diagram combines the time-domain and frequency-domain information and can simultaneously show the frequency content of the signal changing with time, etc.

[0068] In some embodiments, before obtaining the target signal image, the computing system 200 may first obtain the target signal, for example, by intercepting or receiving the target signal. Further, the target signal is converted into a target signal image by the computing system 200 or other devices with image processing capabilities. During the image processing, the device may also perform different operations based on the signal type of the target signal. For example, if the target signal is an analog signal, the analog signal is first digitally processed and converted into a digital signal, and then the digital signal is converted into the target signal image; if the target signal is a digital signal, the target signal image can be directly obtained based on the digital signal.

[0069] As some examples, Figure 4 FIG. 2 shows a schematic diagram of some images obtained by signal conversion provided according to an embodiment of the present specification. As Figure 4 shown, these images are obtained by converting signals of different modulation types. For example, image 410 is obtained by converting a BSKP signal, image 420 is obtained by converting a Costas signal, image 430 is obtained by converting an LFM signal, and image 440 is obtained by converting a P2 signal.

[0070] S320: Input the target signal image into the trained recognition model to identify the target modulation type through the recognition model. The target modulation type is the modulation type of the target signal described by the target signal image.

[0071] In some embodiments, the recognition model can identify one type of image or multiple types of images. For example, recognition model A only supports identifying time-domain diagrams, recognition model B only supports identifying frequency-domain diagrams, and recognition model C can support identifying both time-domain diagrams and frequency-domain diagrams. Those skilled in the art can train or select different recognition models based on the application scenario and requirements. Therefore, the types and categories of images that the recognition model can identify are not limited in the embodiments of this specification.

[0072] Please refer to Figure 3 and Figure 5 , the recognition model at least includes a first feature extraction network and a classification network. The recognition process of the recognition model includes steps S321 - S322:

[0073] S321: Extract first image features from the target signal image through the first feature extraction network.

[0074] Before the recognition model is trained, the first feature extraction network is a network with image understanding ability pre-trained using a first image set. Each image in the first image set is a non-signal image. The first feature extraction network in this specification can be a pre-trained large model with image understanding ability.

[0075] From Figure 4 the shown image, it can be seen that the image obtained by signal conversion is usually a grayscale image with less image detail information. If a feature extraction network with a small parameter scale is used, it may not be able to fully capture the complex features and detail information in the signal, resulting in a decline in recognition performance. Therefore, adopting a first feature extraction network with a large parameter scale can enhance the feature extraction ability on the basis of retaining the global information of the image, enabling the network to extract more abundant and useful features from the grayscale image, and improving the accuracy and robustness of modulation type recognition. The first feature extraction network utilizes the image understanding ability of the large model, and the extracted first image features are more abundant and comprehensive, capable of covering more detail information. In addition, the number of images in the signal field is small and difficult to obtain, while non-signal images (such as images obtained by natural shooting) are easier to obtain compared to images in the signal field. Therefore, pre-training the first feature extraction network with non-signal images can reduce the implementation difficulty of this solution in the signal field.

[0076] S322: Identify the target modulation type based on the first image features through the classification network.

[0077] Since the signal images of different modulation types do not have obvious differences, the effect of directly using the pre-trained first feature extraction network for feature extraction is poor. Therefore, the first feature extraction network trained based on non-signal images cannot be directly used for feature extraction of signal images.

[0078] Therefore, in order to improve the accuracy of signal recognition, the recognition model also needs to be trained. The training process of the recognition model is based on a second image set, and each image in the second image set is a signal image.

[0079] Considering computing power and problem scale, in some embodiments, the recognition model may not be trained from "0". Instead, the sample signal image is used as a new type of image, and the pre-trained first feature extraction network is used to perform feature encoding on the sample signal image. The targeted training of the recognition model based on the sample signal image enables the first feature extraction network to better learn these differences and magnify the differences between signal images of different modulation types through the first feature extraction network. Since the first feature extraction network is pre-trained and already has image understanding ability, during the training process of the recognition model, only a small number of images in the signal field are needed for fine-tuning to generalize the image understanding ability of the first feature extraction network to the signal field, which can improve the training efficiency of the recognition model.

[0080] It should be noted that the number of images in the first image set is greater than / much greater than the number of images in the second image set. The first image set includes non-signal images in multiple scenarios, such as images of people, vehicles, animals, indoor, outdoor, forests, cities, etc. That is to say, the first feature extraction network provided by the present invention can train the model through a larger data set to improve the generalization ability of the network. The large-scale data set can provide more samples for the first feature extraction network, helping the first feature extraction network better understand and learn the features in the images, and improving the generalization ability and robustness of the network.

[0081] The signal modulation type recognition method provided in this specification converts a target signal into an image domain to obtain a target signal image, and uses a recognition model to recognize the target signal image to obtain the target modulation type. Different from the prior art that directly extracts features from a signal in the signal domain and recognizes the signal modulation type based on the extracted features, in this specification, the differences in the signal are visualized through an image, and the features of the signal are automatically extracted from the target signal image, avoiding the limitations and errors of manual analysis and improving the signal recognition accuracy. And by converting the signal into an image form for analysis, the model can identify potential useful signal features in the noise, enhancing the robustness in complex environments. Further, the first feature extraction network in the recognition model is a network with image understanding ability pre-trained using non-signal images (i.e., the images in the first image set), which enables the recognition model to understand the target signal image and extract richer features based on the understood content, thereby improving the accuracy of the recognition result. On this basis, the recognition model is also trained more specifically based on the signal images converted from sample signals (i.e., the images in the second image set), which can enhance the accuracy of the recognition model for signal image recognition while ensuring the generalization ability of the model, and enhance the overall performance ability of the recognition model.

[0082] In some embodiments, the first feature extraction network may include multiple networks and connection layers. The features extracted by different networks in the first feature extraction network are also different, and the first image feature is obtained after different features are fused. The first feature extraction network performs shallow feature extraction and deep feature extraction on the target signal image to obtain the shallow feature and the deep feature corresponding to the target signal image; and the first feature extraction network performs multi-scale feature fusion on the shallow feature and the deep feature to obtain the first image feature.

[0083] Among them, the shallow feature refers to the feature extracted by the shallow network in the first feature extraction network, and the deep feature refers to the feature extracted by the deep network in the first feature extraction network. Generally, the shallow network has a smaller receptive field and can utilize more fine-grained feature information. Therefore, the shallow feature usually focuses on local and low-level features. As the downsampling or the number of convolutions in the deep network increases, the receptive field gradually increases, and the overlapping area between receptive fields also continuously increases. Therefore, the deep feature is the information extracted by the subsequent layers of the feature extraction network, representing the high-level and abstract features of the image. The shallow feature focuses on the local and detailed characteristics of the signal image, while the deep feature abstracts the shallow feature to form a global semantic expression. In the image of a communication signal, the shallow feature can be the local waveform characteristics or texture changes, and the deep feature can be the overall distribution reflecting the modulation mode or the semantic mode of a complex signal. By combining the deep feature and the shallow feature, the first feature extraction network can achieve a comprehensive analysis of the signal image.

[0084] To further improve the accuracy and precision of the recognition model, please refer to Figure 6 , the recognition model further includes a second feature extraction network, and the recognition process of the recognition model further includes: extracting second image features from the target signal image through the second feature extraction network; step S322 further includes recognizing the target modulation type based on the first image features and the second image features through the classification network.

[0085] The first feature extraction network and the second feature extraction network satisfy at least one of the following: the number of network parameters of the first feature extraction network is greater than that of the second feature extraction network, the network depth of the first feature extraction network is greater than that of the second feature extraction network, or the network width of the first feature extraction network is greater than that of the second feature extraction network.

[0086] The differences in the number of network parameters, network depth, or network width between the first feature extraction network and the second feature extraction network enable the first feature extraction network to have stronger expression ability and learning ability in dealing with complex tasks, and can extract more complex and more global features. The second feature extraction network is based on a lightweight model and focuses on lightweight or local feature extraction. By fusing and recognizing the features of the two feature extraction networks with different dimensions through the classification network, the complementarity of global features and local features can be realized, and the feature expression ability of the recognition model in this specification can be enhanced.

[0087] In some embodiments, the second feature network can be a model with a small number of parameters trained for the current signal processing scenario. For example, the second feature extraction network can be a latent net. The latent net can provide a new feature perspective by reducing the dimension or performing a non-linear transformation on the features, supplementing the information that may be missed in the first feature extraction network, increasing the diversity of features, and thus improving the performance of the recognition model.

[0088] In addition to the differences between the first feature network and the second feature network, the differences in the features extracted by the two networks can be reflected as follows: the first image features are the features obtained by understanding the content in the target signal image and encoding the understood content; the second image features are the features obtained by encoding the target signal image from a high-dimensional space to a low-dimensional space.

[0089] By designing the feature encoders of the first feature network and the second feature network differently, the features extracted by the two networks can be made different. The encoder of the first feature extraction network can be designed to capture the global features and complex semantic information in the target signal image, and include more parameters, deeper layers, or a wider network structure, in order to extract high-level patterns and details in the image. The encoder of the second feature extraction network can be designed to be more lightweight, focusing on compressing high-dimensional image information through dimensionality reduction techniques, and only retaining the image features that are most critical for modulation type recognition, in order to reduce computational complexity while enhancing the ability to extract key information.

[0090] The first image feature is extracted through semantic analysis and structural understanding of the content of the target signal image, reflecting the global and complex signal characteristics of the target signal image. For example, it can include but is not limited to in-depth interpretation of spectral distribution and structural information, and encoding this high-level information into features suitable for model recognition. This feature extraction method retains the global characteristics of the signal and can clearly present the complex patterns contained in the modulation type, such as amplitude changes, phase changes, and frequency domain features. The second image feature, on the other hand, aims to remove redundant information through dimensionality reduction processing. By compressing the content of the target signal image, it can then extract the core information in the modulation signal, focusing on the most significant and discriminative parts of the signal, and thus can effectively filter out noise interference.

[0091] By fusing the first image feature and the second image feature in this specification, a deep fusion of global semantic information and core features can be achieved, not only improving the accuracy of the classification of the recognition model, but also enhancing the model's adaptability to complex modulation types.

[0092] The first image feature and the second image feature obtained above also need to be input into a classification network for feature recognition to identify the modulation type. Specifically, the classification network performs alignment processing on the first image feature and the second image feature, making the dimensions of the first image feature and the second image feature equal; the classification network performs fusion processing on the aligned first image feature and the second image feature to obtain a fusion feature; the classification network performs classification processing on the fusion feature to obtain the target modulation type.

[0093] Since the first image feature and the second image feature are based on features of different dimensions, the classification network needs to first align the dimensions of the two types of image features to ensure the consistency of the feature expression space, thereby eliminating the mismatch problem caused by feature differences.

[0094] The classification network also needs to perform a fusion process on the aligned features, combining the global characteristics with the key attributes to form a more comprehensive fused feature representation, including: determining the first weight corresponding to the first image feature and determining the second weight corresponding to the second image feature; and performing a weighted calculation on the first image feature and the second image feature according to the first weight and the second weight to obtain the fused feature.

[0095] In some embodiments, the importance of the first image feature and the second image feature can be quantified using an attention mechanism or an adaptive learning algorithm to dynamically generate the first weight and the second weight. The weighted calculation of the features can be completed by element-wise addition or concatenation operations after the dot product of the feature vectors to generate the fused feature.

[0096] Furthermore, the classification network can continue to optimize the fused feature, for example, by using a fully connected layer to improve the recognition ability. The fused feature can also be mapped to the category space through a fully connected layer to obtain the category scores of the fused feature for different modulation types. The category scores represent the matching degree of the fused feature with each modulation type, and the higher the value, the greater the possibility of that modulation type. Further, these category scores can be converted into a category probability distribution through an activation function, and the target modulation type can be identified based on the category probability distribution. Through the conversion of the above category scores and probability distribution, not only the accuracy of modulation type recognition is improved, but also the interpretability of the model is enhanced.

[0097] In the above embodiments, the classification network can optimize the efficiency of feature fusion through alignment, reducing the waste of computing resources. The recognition ability is enhanced after the fusion of two different-dimensional features for recognition in a high-noise environment. Moreover, the classification network provided in this specification can also adjust the contribution ratio of features according to the scenario, enhancing the adaptability and recognition performance of the recognition model.

[0098] In some embodiments, the classification network is any one of the following: an encoder-decoder network based on self-attention mechanism (Transformer); an encoder-decoder network with skip connections (U-Net); or a multi-layer perceptron network (Multilayer Perceptron Network, MLPs).

[0099] Among them, the Transformer network uses the self-attention mechanism to model the long-range dependencies in the input features, automatically combines the first image feature and the second image feature, better recognizes features, and is suitable for processing complex signal patterns. U-Net adopts a symmetric encoder-decoder structure and fuses shallow and deep features through skip connections. It is suitable for tasks that require multi-scale feature representation, such as signal classification in high-noise environments or radar image recognition. MLPs can learn the mapping relationship between features and modulation categories through multiple layers of linear transformations and non-linear activation functions. The structure of MLPs is simple and computationally efficient, and it is suitable for scenarios with low input feature dimensions or low task complexity, such as the fast classification task of standard modulation types.

[0100] In actual use, an appropriate classification network architecture can be selected for training according to requirements. For example, Transformer is suitable for processing complex and diverse signal characteristics, U-Net improves the robustness to high-noise signals through multi-scale feature fusion, while MLPs provide a lightweight classification solution. By flexibly selecting the network architecture, the actual needs of different signal scenarios can be met, and the flexibility of the modulation type recognition usage scenario can be enhanced.

[0101] Based on the modulation type recognized by the recognition system 120 in the above embodiments, if the target signal is a signal emitted by a radar, the recognition system 120 can also determine the type of the radar based on the target modulation type. If the target signal is a communication signal, the recognition system 120 can also demodulate and analyze the content of the target signal based on the target modulation type.

[0102] In the radar signal scenario, by analyzing the target modulation type, the recognition system 120 can infer the specific type of the radar. Based on the specific type, it can be used for tasks such as target detection, trajectory tracking, and radar signal source identification. For example, in a complex electromagnetic environment, the device can infer the radar type through the modulation type of the intercepted radar signal, identify whether the signal source is a friendly device, and assist in electronic countermeasures and radar interference optimization.

[0103] In the communication signal scenario, the recognition system 120 can demodulate and analyze the signal content using the recognized modulation type. For example, infer the coding method, information structure, or communication protocol of the signal based on the modulation type. For common modulation types (such as QPSK, 16QAM, etc.), the device can further analyze the data packets in the signal, decode the information content, and identify its purpose (such as voice, video, or data transmission).

[0104] Figure 7Also shown is a flowchart of a training method P400 for an identification model provided according to an embodiment of this specification. Before training, it is necessary to first determine the specific dimensions, number of layers, etc. of each module in the identification model. The identification model in the embodiments of this specification at least includes a first feature extraction network and a classification network. As mentioned before, the training system 110 can execute the training method P400 of the identification model in this specification. Specifically, the processor 220 in the training system 110 can read the instruction set stored in its local storage medium, and then execute the training method P400 of the identification model in this specification according to the provisions of the instruction set. As Figure 7 shown, the method P400 may include steps S410 - S430:

[0105] S410: Obtain an identification model to be trained, where the identification model at least includes a first feature extraction network and a classification network, and the first feature extraction network is a network with image understanding ability pre-trained using a first image set, and each image in the first image set is a non-signal image.

[0106] S420: Obtain a second image set, where the second image set includes a plurality of sample signal images and the annotation information corresponding to each sample signal image, and the annotation information represents the actual modulation type adopted by the sample signal described by the sample signal image.

[0107] The second image set can be obtained by converting the sample signals in the existing public dataset, converting the one-dimensional signal data in the dataset into a multi-dimensional image representation. Before training, preprocessing such as data augmentation can be performed on the second image set, for example, rotation, flipping, random masking, etc.

[0108] Furthermore, each sample signal image is accompanied by corresponding annotation information, which represents the actual modulation type adopted by the original signal. When making annotations, the model ability can be optimized using simulation data first, and then based on the optimized model, rough labeling can be performed on more real data (unknown labels) and then manual fine labeling can be added, and then added to the model for learning to continuously enhance the practicality of the identification model.

[0109] To ensure the diversity and applicability of the image set, the sample signals of each modulation type can include multiple signal-to-noise ratios, and the sample signals include radar signals and communication signals. By introducing sample signals with multiple signal-to-noise ratios, it helps to improve the generalization ability and robustness of the trained identification model, enabling it to adapt to different actual application scenarios.

[0110] S430: Use the second image set to perform multiple iterative trainings on the identification model to obtain the trained identification model.

[0111] Wherein, the process of each iterative training includes steps S431 - S433:

[0112] S431: Extract first image features from the sample signal image through the first feature extraction network.

[0113] S432: Identify the predicted modulation type corresponding to the sample signal based on the first image features through the classification network.

[0114] S433: Update the parameters of the recognition model with the training objective of minimizing the difference between the predicted modulation type and the actual modulation type.

[0115] In some embodiments, step S433 includes: determining the error between the predicted modulation type and the actual modulation type; determining the gradient of the error with respect to the parameters of the recognition model, and adjusting the parameters of the recognition model based on the gradient.

[0116] First, determine the error between the predicted modulation type and the actual modulation type through a loss function. Then, based on the error, calculate the gradient of the loss function with respect to the model parameters. Among them, the gradient reflects the information about the current parameter adjustment direction and amplitude, and can guide the model to optimize in the direction of reducing the error. Next, the gradient is propagated back through each parameter layer of the recognition model to adjust the model weights and biases layer by layer. The above process is iterated continuously to gradually optimize the parameters of the recognition model. As the training progresses, the prediction results of the model will be closer to the actual modulation type, thereby achieving a high-accuracy classification performance. By precisely quantifying the error between the prediction and the actual, the model can effectively handle complex signal patterns and diverse modulation types. Finally, ablation experiments can also be conducted on the network structure of the recognition model, testing different modules of the recognition model to determine the rationality of the network design.

[0117] In some embodiments, the recognition model further includes a second feature extraction network, and the process of each iterative training further includes: extracting second image features from the sample signal image through the second feature extraction network; step S432 further includes: identifying the predicted modulation type based on the first image features and the second image features through the classification network.

[0118] In some embodiments, the first feature extraction network and the second feature extraction network satisfy at least one of the following: the number of network parameters of the first feature extraction network is greater than that of the second feature extraction network, the network depth of the first feature extraction network is greater than that of the second feature extraction network, or the network width of the first feature extraction network is greater than that of the second feature extraction network.

[0119] In the embodiments of this specification, selecting a large model with a large number of network parameters as the base model of the first feature extraction network can solve the problem of the low generalization degree of traditional image recognition based on small models. These models have fewer parameters and are difficult to capture rich information in images. In contrast, large models with a large number of parameters have more parameters and can better capture features in different images, improving the generalization ability of image recognition. Specifically, the number of parameters of the first feature extraction network can be increased by increasing the depth and width of the network, thereby improving the representation ability of the model and the generalization ability of feature extraction. This can better handle complex image features, improve the recognition accuracy of the recognition model for different scenarios and samples, and reduce the overfitting risk of the recognition model in specific scenarios. In addition, the generalization ability of the model can also be improved by training the model with a larger dataset. The large-scale dataset can provide more samples for the model, helping the model better understand and learn the features in the images, and improving the generalization ability and robustness of the model.

[0120] In some embodiments, the first image feature is a feature obtained by understanding the content in the sample signal image and encoding the understood content; the second image feature is a feature obtained by encoding the sample signal image from a high-dimensional space to a low-dimensional space.

[0121] In some embodiments, the step of identifying the predicted modulation type by the classification network based on the first image feature and the second image feature includes: performing an alignment process on the first image feature and the second image feature by the classification network to make the dimensions of the first image feature and the second image feature equal; performing a fusion process on the aligned first image feature and the second image feature by the classification network to obtain a fusion feature; and performing a classification process on the fusion feature by the classification network to obtain the predicted modulation type.

[0122] In some embodiments, performing a fusion process on the aligned first image feature and the second image feature by the classification network to obtain a fusion feature includes: determining a first weight corresponding to the first image feature and a second weight corresponding to the second image feature; and performing a weighted calculation on the first image feature and the second image feature according to the first weight and the second weight to obtain the fusion feature.

[0123] In some embodiments, extracting first image features from the sample signal image through the first feature extraction network includes: extracting shallow features and deep features of the sample signal image through the first feature extraction network, and performing multi-scale feature fusion on the shallow features and the deep features through the first feature extraction network to obtain the first image features.

[0124] The operations in the process of training the recognition model described above are similar to other specific operations in the process of signal recognition based on the recognition model, which will not be elaborated in this specification. The difference is that the input in the training process is the sample signal image corresponding to the sample signal.

[0125] The recognition model trained based on the above network structure has higher running efficiency and lower computational cost compared with traditional large models. This means that in the process of signal recognition, model training and inference can be carried out more quickly, thus meeting the real-time requirements. The trained recognition model has good accuracy and generalization ability, which means that even under limited resource conditions, a relatively high level of signal recognition effect can be achieved. The trained recognition model can also be applied to signal recognition in resource-constrained environments, and can be applied to scenarios such as embedded systems and mobile devices to meet the needs of practical engineering applications. The recognition model based on the combination of the feature extraction network and the classification network can give full play to the architectural advantages of each network.

[0126] On the other hand, this specification provides a computer-readable non-transitory storage medium storing at least one instruction set for identifying the signal modulation type. When the at least one instruction set is executed by a processor, the at least one instruction set guides the processor to implement the steps of the signal modulation type recognition method P300 described in this specification.

[0127] On the other hand, this specification provides a computer-readable non-transitory storage medium storing at least one instruction set for training the recognition model. When the at least one instruction set is executed by a processor, the at least one instruction set guides the processor to implement the steps of the recognition model training method P400 described in this specification.

[0128] In some possible embodiments, various aspects of this specification can also be implemented in the form of a program product, which includes program code. When the program product runs on the computing system 200, the program code is used to cause the computing system 200 to execute the steps of the signal modulation type recognition method P300 described in this specification, or execute the steps of the recognition model training method P400 described in this specification. The program product for implementing the above methods can be a portable compact disc read-only memory (CD-ROM) including program code and can run on the computing system 200. However, the program product of this specification is not limited to this. In this specification, the readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system. The program product can adopt any combination of one or more readable media. The readable media can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, but not be limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the readable storage medium include: portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. The computer-readable storage medium can include a data signal propagated as part of a carrier wave in a baseband, where the readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than the readable storage medium, and this readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above. The program code for performing the operations of this specification can be written in any combination of one or more programming languages, including object-oriented programming languages - such as Java, C++, etc., and also including conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the computing system 200, partially on the computing system 200, executed as an independent software package, partially on the computing system 200 and partially on a remote electronic device, or entirely on a remote electronic device.

[0129] The above description has been made of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require a particular or sequential order to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] In summary, after reading this detailed disclosure, those skilled in the art will appreciate that the foregoing detailed disclosure may be presented by way of example only and is not necessarily limiting. Although not explicitly stated herein, those skilled in the art will understand that this specification is intended to encompass various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be proposed by this specification and are within the spirit and scope of the exemplary embodiments of this specification.

[0131] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean that the specific features, structures, or characteristics described in connection with that embodiment may be included in at least one embodiment of this specification. Thus, it should be emphasized and understood that two or more references to "an embodiment" or "one embodiment" or "alternative embodiments" in various parts of this specification do not necessarily all refer to the same embodiment. Additionally, the specific features, structures, or characteristics may be appropriately combined in one or more embodiments of this specification.

[0132] It should be understood that in the foregoing description of the embodiments of this specification, for the purpose of helping to understand a feature, and for the purpose of simplifying this specification, this specification combines various features in a single embodiment, figure, or its description. However, this does not mean that the combination of these features is necessary, and those skilled in the art may well mark out some of the devices as separate embodiments when reading this specification. That is to say, the embodiments in this specification can also be understood as the integration of multiple sub - embodiments. And the content of each sub - embodiment is also valid when it has fewer features than all the features of a single foregoing disclosed embodiment.

[0133] Each patent, patent application, publication of a patent application, and other materials cited herein, such as articles, books, specifications, publications, documents, items, etc., except those inconsistent or conflicting with this document or those having a limiting effect on the broadest scope of the claims, may be incorporated herein by reference and used for all purposes now or hereafter associated with this document. Further, in the event of any inconsistency or conflict between the description, definition, and / or use of relevant terms in any material and the description, definition, and / or use of such terms in this document, the terms in this document shall control.

[0134] Finally, it should be understood that the embodiments of the application disclosed herein are illustrative of the principles of the embodiments of this specification. Other modified embodiments are also within the scope of this specification. Accordingly, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art may implement the application in this specification by adopting alternative configurations based on the embodiments in this specification. Therefore, the embodiments of this specification are not limited to the embodiments precisely described in the application.

Claims

1. A method for identifying a signal modulation type, comprising: Obtaining a target signal image; Inputting the target signal image into a trained recognition model to identify a target modulation type through the recognition model, where the target modulation type is the modulation type of the target signal described by the target signal image. Among them, the recognition model at least includes a first feature extraction network and a classification network, and the recognition process of the recognition model includes: Extracting first image features from the target signal image through the first feature extraction network, and identifying the target modulation type based on the first image features through the classification network. Before the recognition model is trained, the first feature extraction network is a network with image understanding ability pre-trained using a first image set, and each image in the first image set is a non-signal image. The training process of the recognition model is based on a second image set, and each image in the second image set is a signal image.

2. The method according to claim 1, wherein, The recognition model further includes a second feature extraction network, and the recognition process of the recognition model further includes: extracting second image features from the target signal image through the second feature extraction network; Identifying the target modulation type based on the first image features through the classification network includes: Identifying the target modulation type based on the first image features and the second image features through the classification network.

3. The method according to claim 2, wherein, The first feature extraction network and the second feature extraction network satisfy at least one of the following: The number of network parameters of the first feature extraction network is greater than that of the second feature extraction network. The network depth of the first feature extraction network is greater than that of the second feature extraction network, or The network width of the first feature extraction network is greater than that of the second feature extraction network.

4. The method according to claim 2, wherein The first image features are features obtained by understanding the content in the target signal image and encoding the understood content. The second image features are features obtained by encoding the target signal image from a high-dimensional space to a low-dimensional space.

5. The method according to claim 2, wherein Identifying the target modulation type based on the first image features and the second image features through the classification network includes: Performing alignment processing on the first image features and the second image features through the classification network to make the dimensions of the first image features and the second image features equal; Performing fusion processing on the aligned first image features and second image features through the classification network to obtain fused features; Performing classification processing on the fused features through the classification network to obtain the target modulation type.

6. The method according to claim 5, wherein, Performing fusion processing on the aligned first image features and second image features through the classification network to obtain fused features, including: Determining a first weight corresponding to the first image features and a second weight corresponding to the second image features; and Performing weighted calculation on the first image features and the second image features according to the first weight and the second weight to obtain the fused features.

7. The method according to claim 1, wherein Performing feature extraction on the target signal image through the first feature extraction network to obtain first image features, including: Performing shallow feature extraction and deep feature extraction on the target signal image through the first feature extraction network to obtain shallow features and deep features corresponding to the target signal image; and Performing multi-scale feature fusion on the shallow features and the deep features through the first feature extraction network to obtain the first image features.

8. The method according to claim 1, wherein The classification network is any one of the following: An encoder-decoder network based on a self-attention mechanism; An encoder-decoder network with skip connections; or A multi-layer perceptron network.

9. The method according to claim 1, wherein The target signal is a signal emitted by a radar, and the method further includes: Determining the type of the radar based on the target modulation type.

10. The method according to claim 1, wherein, The target signal is a communication signal, and the method further includes: Demodulating and analyzing the content of the target signal based on the target modulation type.

11. A training method for an identification model, including: Obtaining an identification model to be trained, where the identification model at least includes a first feature extraction network and a classification network, and the first feature extraction network is a network with image understanding ability pre-trained using a first image set, and each image in the first image set is a non-signal image; Obtaining a second image set, where the second image set includes a plurality of sample signal images and annotation information corresponding to each sample signal image, and the annotation information represents the actual modulation type adopted by the sample signal described by the sample signal image; And Performing multiple iterative trainings on the identification model using the second image set to obtain a trained identification model, where the process of each iterative training includes: Performing feature extraction on the sample signal image through the first feature extraction network to obtain first image features, Identifying the predicted modulation type corresponding to the sample signal through the classification network based on the first image features, and Taking minimizing the difference between the predicted modulation type and the actual modulation type as the training objective, Updating the parameters of the identification model.

12. The method according to claim 11, wherein, The identification model further includes a second feature extraction network, and the process of each iterative training further includes: performing feature extraction on the sample signal image through the second feature extraction network to obtain second image features; Identifying the predicted modulation type corresponding to the sample signal through the classification network based on the first image features, including: Identifying the predicted modulation type through the classification network based on the first image features and the second image features.

13. The method according to claim 12, wherein, The first feature extraction network and the second feature extraction network satisfy at least one of the following: The number of network parameters of the first feature extraction network is greater than the number of network parameters of the second feature extraction network, The network depth of the first feature extraction network is greater than the network depth of the second feature extraction network, or The network width of the first feature extraction network is greater than the network width of the second feature extraction network.

14. The method according to claim 12, wherein, The first image features are features obtained by understanding the content in the sample signal image and encoding the understood content; The second image feature is a feature obtained by encoding the sample signal image from a high-dimensional space to a low-dimensional space.

15. The method according to claim 12, wherein, The identifying the predicted modulation type by the classification network based on the first image feature and the second image feature includes: performing an alignment process on the first image feature and the second image feature by the classification network to make the dimensions of the first image feature and the second image feature equal; performing a fusion process on the aligned first image feature and second image feature by the classification network to obtain a fusion feature; performing a classification process on the fusion feature by the classification network to obtain the predicted modulation type.

16. The method according to claim 15, wherein Performing a fusion process on the aligned first image feature and second image feature by the classification network to obtain a fusion feature includes: determining a first weight corresponding to the first image feature and determining a second weight corresponding to the second image feature; and performing a weighted calculation on the first image feature and the second image feature according to the first weight and the second weight to obtain the fusion feature.

17. The method according to claim 11, wherein, Obtaining a first image feature by performing feature extraction on the sample signal image by the first feature extraction network includes: performing shallow feature extraction and deep feature extraction on the sample signal image by the first feature extraction network to obtain a shallow feature and a deep feature corresponding to the sample signal image; and performing multi-scale feature fusion on the shallow feature and the deep feature by the first feature extraction network to obtain the first image feature.

18. The method according to claim 11, wherein Updating the parameters of the recognition model with the training objective of minimizing the difference between the predicted modulation type and the actual modulation type includes: determining the error between the predicted modulation type and the actual modulation type; determining the gradient of the error with respect to the parameters of the recognition model and adjusting the parameters of the recognition model based on the gradient.

19. A system for identifying a signal modulation type, comprising: at least one storage medium storing at least one instruction set for identifying a signal modulation type; and at least one processor communicatively connected to the at least one storage medium, wherein when the system for identifying a signal modulation type runs, the at least one processor reads the at least one instruction set and executes the method for identifying a signal modulation type according to any one of claims 1-10 based on the indication of the at least one instruction set.

20. A training system for a recognition model, comprising: at least one storage medium storing at least one instruction set for training a recognition model; and at least one processor communicatively connected to the at least one storage medium, wherein when the training system for a recognition model runs, the at least one processor reads the at least one instruction set and executes the method for training a recognition model according to any one of claims 11-18 based on the indication of the at least one instruction set.

Images