SAR anti-deception interference method, device and equipment based on deep learning network

Abstract

The application provides a SAR anti-deception interference method and device based on a deep learning network and equipment, and relates to the technical field of data analysis. In the application, first, the original echo data received by the SAR receiver is processed to form an original SAR complex image; second, the semantic information of the original SAR complex image is encoded and enhanced based on the semantic information in the original echo data and the data processed by at least one node before the third node in the target processing, to form a complex image enhanced semantic vector; then, the complex image enhanced semantic vector is decoded to obtain a deception interference area identification result; finally, based on the deception interference area identification result, the deception interference area in the original SAR complex image is interfered and suppressed to obtain an anti-deception interference SAR complex image. Based on the above, the problem that the reliability of the anti-deception interference in the prior art is relatively low can be improved.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and more specifically, to a SAR anti-spoofing interference method, apparatus, and device based on deep learning networks. Background Technology

[0002] Synthetic Aperture Radar (SAR) is an active microwave imaging remote sensing device with all-weather, all-day, high-resolution imaging capabilities, widely used in topographic mapping, environmental monitoring, and other fields. With the widespread application of SAR technology in various fields, the electromagnetic environment it faces is becoming increasingly complex. Current jamming techniques against SAR are mainly divided into two categories: noise suppression jamming and deception jamming. Noise suppression jamming reduces the signal-to-noise ratio of the SAR receiver by transmitting broadband noise signals, but it is easily detected and suppressed. Deception jamming intercepts the SAR transmitted signal through a digital radio frequency memory (DRFM), modulates it, and then forwards it, creating false targets in the SAR imaging results that resemble the real target, thus possessing higher concealment. Traditional machine learning-based methods can identify deception jamming, such as using support vector machines and random forests to distinguish between real targets and jamming signals, but feature engineering is complex and its generalization ability to complex scenarios is insufficient. In particular, when the power of the jamming signal is close to or exceeds that of the real target echo, it is difficult to effectively distinguish between the jamming and the real target. In other words, existing technologies suffer from relatively low reliability in resisting deception jamming. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a SAR anti-spoofing jamming method, apparatus and device based on deep learning networks, so as to improve the problem of relatively low reliability of anti-spoofing jamming in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: A SAR anti-spoofing jamming method based on deep learning networks includes: The raw echo data received by the SAR receiver is subjected to target processing to form a raw SAR complex image. The target processing includes at least the range compression processing of the first node, the azimuth compression processing of the second node, and the complex image generation processing of the third node performed sequentially. Using the coding model in the anti-spoofing interference identification network formed by deep learning, the semantic information of the original SAR complex image is encoded and enhanced based on the semantic information in the original echo data and the data processed by at least one node before the third node in the target processing, forming a complex image enhanced semantic vector. Using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the complex image is decoded to obtain the identification result of the spoofing interference region; Based on the identification results of the deception interference region, interference suppression is performed on the deception interference region in the original SAR complex image to obtain an anti-deception interference SAR complex image.

[0005] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of encoding and enhancing the semantic information of the original SAR complex image based on the semantic information in the original echo data and the processed data of at least one node before the third node in the target processing, using the coding model in the anti-spoofing interference identification network formed by deep learning, to form a complex image enhanced semantic vector, includes: The original echo data, the data processed by at least one node before the third node in the target processing, and the original SAR complex image are respectively loaded into the coding model in the anti-deception interference recognition network formed based on deep learning. The raw echo data is semantically encoded to form a raw echo semantic vector; Semantic encoding is performed on the data processed by at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector. The original SAR complex image is semantically encoded to form a complex image semantic vector; In the deep encoding process of the complex image semantic vector, the encoding depth of the complex image semantic vector is controlled by the original echo semantic vector, and the encoding direction of the complex image semantic vector is controlled by the at least one intermediate data semantic vector, so as to obtain the complex image enhanced semantic vector through deep encoding.

[0006] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of controlling the encoding depth of the complex image semantic vector through the original echo semantic vector and controlling the encoding direction of the complex image semantic vector through at least one intermediate data semantic vector, so as to obtain the complex image enhanced semantic vector through deep encoding, includes: Based on the embedding vectors corresponding to the target signal power, noise floor power, and signal-to-noise ratio in the original echo data, the original echo semantic vector is focused and mined to form an interference level characterization vector. Based on the interference level characterization vector, a mapping output is performed to obtain the target depth, wherein the target depth is used to indicate the encoding depth of the complex image semantic vector. According to the encoding direction represented by the at least one intermediate data semantic vector, the complex image semantic vector is depth encoded corresponding to the target depth to form a complex image enhanced semantic vector.

[0007] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of performing depth encoding on the complex image semantic vector according to the encoding direction represented by the at least one intermediate data semantic vector to form a complex image enhanced semantic vector includes: Perform semantic space transformation on the at least one intermediate data semantic vector to form a transformed data semantic vector; In the depth encoding of the first depth, the complex image semantic vector is weighted and residually connected based on the focus direction representation parameter of the transformed data semantic vector or the focus direction representation parameter of the related semantic representation between the transformed data semantic vector and the complex image semantic vector to obtain the complex image depth semantic vector of the first depth. In each depth encoding between the second depth and the target depth, the complex image depth semantic vector of the previous depth is weighted and residually connected based on the focus direction representation parameter represented by the focus vector of the transformed data semantic vector at the corresponding depth or the focus direction representation parameter based on the relevant semantic representation between the focus vector of the transformed data semantic vector at the corresponding depth and the complex image depth semantic vector of the previous depth, so as to obtain the complex image depth semantic vector of the corresponding depth. The depth semantic vector of the complex image at the target depth is determined as the complex image enhancement semantic vector.

[0008] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of semantically encoding the original echo data to form an original echo semantic vector includes: The original echo data is convolved to form the original echo convolution vector; The original echo convolution vector is pooled and compressed to form an original echo pooling vector, and the original echo convolution vector is segmented to form multiple original echo segmentation vectors. Based on each of the original echo segmentation vectors, the original echo pooling vectors are focused to form each corresponding original echo focusing vector; By aggregating each of the original echo focusing vectors, the original echo semantic vector is obtained.

[0009] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of semantically encoding the data processed by at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector includes: The azimuth-compressed data obtained by the second node is convolved to form an azimuth-compressed convolution vector. The azimuth compressed convolution vector is pooled to form an azimuth compressed pooling vector, and the azimuth compressed convolution vector is segmented to form multiple azimuth compressed segmentation vectors. Based on each of the azimuth compression segmentation vectors, the azimuth compression pooling vectors are focused to form corresponding azimuth compression focusing vectors; By aggregating each of the aforementioned orientation compression focus vectors, an intermediate data semantic vector is obtained.

[0010] In a preferred embodiment of this application, in the aforementioned SAR anti-spoofing interference method based on deep learning networks, the step of semantically encoding the original SAR complex image to form a complex image semantic vector includes: The original SAR complex image is convolved to form a complex image convolution vector; The complex image convolution vector is pooled to form a complex image pooling vector, and the complex image convolution vector is segmented to form multiple complex image segmentation vectors; Based on each of the complex image segmentation vectors, the complex image pooling vectors are focused to form a corresponding complex image focusing vector; By aggregating the focus vectors of each complex image, a complex image semantic vector is obtained.

[0011] In a preferred embodiment of this application, the aforementioned SAR anti-spoofing jamming method based on deep learning networks further includes: The original echo data of the sample is processed to form the original SAR complex image of the sample. Using the coding model in the initial anti-spoofing interference identification network, based on the semantic information in the original echo data of the sample and the data processed by at least one node before the third node in the target processing, the semantic information of the original SAR complex image of the sample is encoded and enhanced to form a sample complex image enhanced semantic vector. Using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the sample complex image is decoded to obtain the corresponding spoofing interference region recognition result; Based on the error between the spoofing interference region identification result corresponding to the original echo data of the sample and the spoofing interference region label corresponding to the original echo data of the sample, the anti-spoofing interference identification network is trained to obtain the trained anti-spoofing interference identification network.

[0012] This application also provides a SAR anti-spoofing jamming device based on a deep learning network, comprising: The echo data processing module is used to perform target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image. The target processing includes at least the range compression processing of the first node, the azimuth compression processing of the second node, and the complex image generation processing of the third node performed sequentially. The data encoding module is used to encode and enhance the semantic information of the original SAR complex image based on the semantic information in the original echo data and the data processed by at least one node before the third node in the target processing, using the encoding model in the anti-spoofing interference recognition network formed by deep learning, to form a complex image enhanced semantic vector. The semantic decoding module is used to decode the enhanced semantic vector of the complex image using the decoding model in the anti-spoofing interference recognition network to obtain the identification result of the spoofing interference region; The interference suppression module is used to suppress the deception interference region in the original SAR complex image based on the deception interference region identification result, so as to obtain an anti-deception interference SAR complex image.

[0013] Based on the above, this application also provides an electronic device, including: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-described SAR anti-spoofing jamming method based on deep learning networks.

[0014] The SAR anti-spoofing interference method, apparatus, and device based on deep learning networks provided in this application firstly perform target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image. Secondly, based on the semantic information in the raw echo data and the data processed by at least one node before the third node in the target processing, the semantic information of the raw SAR complex image is encoded and enhanced to form a complex image enhanced semantic vector. Then, the complex image enhanced semantic vector is decoded to obtain the deception interference region identification result. Finally, based on the deception interference region identification result, interference suppression is performed on the deception interference region in the raw SAR complex image to obtain an anti-spoofing interference SAR complex image. Based on the above, since the encoding process not only mines the semantic information in the formed raw SAR complex image but also enhances the semantics based on the raw echo data and intermediate processed data, the semantic representation accuracy of the formed complex image enhanced semantic vector is higher. Therefore, the reliability of the identification result obtained from decoding is higher, thereby ensuring the reliability of interference suppression. Thus, the solution provided in this application can improve the problem of relatively low reliability in anti-spoofing interference in existing technologies. Attached Figure Description

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0016] Figure 1 A structural block diagram of an electronic device provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating the SAR anti-spoofing interference method based on deep learning networks provided in the embodiments of this application.

[0018] Figure 3 A schematic diagram illustrating the encoding enhancement provided in an embodiment of this application.

[0019] Figure 4 This is a schematic diagram of depth encoding provided in an embodiment of this application.

[0020] Figure 5 This is a block diagram of a SAR anti-spoofing jamming device based on a deep learning network provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] like Figure 1 As shown in the illustration, this application provides an electronic device. The electronic device may include a memory, a processor, and a SAR anti-spoofing jamming device based on a deep learning network.

[0024] Specifically, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The SAR anti-spoofing jamming device based on deep learning networks includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the SAR anti-spoofing jamming device based on deep learning networks, to implement the SAR anti-spoofing jamming method based on deep learning networks provided in this application embodiment.

[0025] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0026] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0027] Understandable Figure 1 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices.

[0028] Combination Figure 2 This application also provides a SAR anti-spoofing jamming method based on a deep learning network, applicable to the aforementioned electronic device. The method steps defined in the process of the SAR anti-spoofing jamming method based on a deep learning network can be implemented by the electronic device. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0029] Step S110: Target processing is performed on the raw echo data received by the SAR receiver to form the raw SAR complex image.

[0030] In this embodiment, the electronic device can perform target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image. The target processing includes at least three sequential steps: range compression processing at a first node, azimuth compression processing at a second node, and complex image generation processing at a third node. Specifically, the raw echo data is first subjected to range compression processing to obtain range-compressed data; then, the range-compressed data is subjected to azimuth compression processing to obtain azimuth-compressed data; and finally, a raw SAR complex image is generated based on the azimuth-compressed data.

[0031] Step S120: Using the coding model in the anti-spoofing interference identification network formed by deep learning, the semantic information of the original SAR complex image is encoded and enhanced based on the semantic information in the original echo data and the data processed by at least one node before the third node in the target processing, forming a complex image enhanced semantic vector.

[0032] In this embodiment, after obtaining the original SAR complex image through target processing, the electronic device can utilize the encoding model in an anti-spoofing interference identification network based on deep learning to encode and enhance the semantic information of the original SAR complex image based on the semantic information in the original echo data and the data processed by at least one node before the third node in the target processing, forming a complex image enhanced semantic vector. The anti-spoofing interference identification network is a neural network that can learn the mapping relationship between corresponding samples and labels.

[0033] Step S130: Using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the complex image is decoded to obtain the spoofing interference region recognition result.

[0034] In this embodiment of the application, after encoding the complex image enhanced semantic vector, the electronic device can use the decoding model in the anti-spoofing interference recognition network to decode the complex image enhanced semantic vector and obtain the spoofing interference region recognition result.

[0035] Step S140: Based on the identification result of the deception interference region, the deception interference region in the original SAR complex image is suppressed to obtain an anti-deception interference SAR complex image.

[0036] In this embodiment of the application, after obtaining the deception interference region identification result, the electronic device can suppress the deception interference region in the original SAR complex image based on the deception interference region identification result to obtain an anti-deception interference SAR complex image.

[0037] Based on the above, since the encoding process not only extracts semantic information from the original SAR complex image but also enhances the semantics based on the original echo data and intermediate processed data, the semantic representation accuracy of the enhanced semantic vector of the complex image is higher. Therefore, the reliability of the recognition result obtained from decoding is higher, thereby ensuring the reliability of interference suppression. Thus, the solution provided in this application can improve the relatively low reliability of anti-spoofing interference in existing technologies.

[0038] Firstly, regarding step S110, it should be noted that the specific method of target processing for the raw echo data received by the SAR receiver is not limited and can be selected according to actual needs. For example, in an alternative implementation, the raw echo data received by the SAR receiver can be directly obtained from the SAR receiver, and various processing sub-steps in the target processing can be performed sequentially, such as range compression processing (converting the echo signal from the time domain to the range frequency domain and performing matched filtering), azimuth compression processing (performing azimuth-directed FFT processing on the range-compressed data), and generating a complex image (the specific method can refer to relevant existing technologies).

[0039] Secondly, regarding step S120, it should be noted that the specific method for encoding and enhancing the semantic information of the original SAR complex image is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to improve the accuracy of encoding and enhancement and make the semantic representation accuracy of the resulting complex image enhanced semantic vector higher, the above-mentioned step S120 may further include steps S121, S122, S123, S124, and S125, wherein the details of each step are as follows.

[0040] Step S121: Load the original echo data, the data processed by at least one node before the third node in the target processing, and the original SAR complex image into the coding model in the anti-spoofing interference recognition network formed by deep learning, respectively.

[0041] In this embodiment of the invention, combined with Figure 3 The original echo data, the data processed by at least one node before the third node in the target processing (such as the range compressed data obtained by the range compression processing of the first node, the azimuth compressed data obtained by the azimuth compression processing of the second node), and the original SAR complex image can be loaded into the coding model in the anti-spoofing interference identification network formed by deep learning, respectively. In this way, subsequent semantic coding and coding enhancement (deep coding) can be performed in the coding model.

[0042] Step S122: Semantic encoding is performed on the original echo data to form an original echo semantic vector.

[0043] In this embodiment of the application, after loading the original echo data, the original echo data can be semantically encoded to form an original echo semantic vector. It should be noted that semantic encoding can refer to mining the potential semantic information in the original echo data and representing it in the form of a vector (which can be one-dimensional, two-dimensional, or three-dimensional, etc.), thus obtaining the original echo semantic vector.

[0044] Step S123: Semantically encode the data processed by at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector.

[0045] In this embodiment, after loading the processed data from at least one node before the third node in the target processing, semantic encoding can be performed on the processed data from at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector. It should be noted that semantic encoding can refer to mining the latent semantic information in the processed data from at least one node before the third node in the target processing and representing it in vector form, thus obtaining the intermediate data semantic vector. For example, semantic encoding of the azimuth-compressed data obtained by performing azimuth compression processing on the second node can obtain the first intermediate data semantic vector, and semantic encoding of the azimuth-compressed data obtained by performing distance compression processing on the first node can obtain the second intermediate data semantic vector.

[0046] Step S124: Semantic encoding is performed on the original SAR complex image to form a complex image semantic vector.

[0047] In this embodiment of the application, after loading the original SAR image, the original SAR complex image can be semantically encoded to form a complex image semantic vector. It should be noted that semantic encoding refers to extracting the latent semantic information in the original SAR complex image and representing it in vector form, thus obtaining the complex image semantic vector.

[0048] Step S125: In the deep encoding process of the complex image semantic vector, the encoding depth of the complex image semantic vector is controlled by the original echo semantic vector, and the encoding direction of the complex image semantic vector is controlled by the at least one intermediate data semantic vector, so as to obtain the complex image enhanced semantic vector through deep encoding.

[0049] In this embodiment, after obtaining the original echo semantic vector, the at least one intermediate data semantic vector, and the complex image semantic vector, the encoding depth of the complex image semantic vector can be controlled by the original echo semantic vector during the depth encoding process, and the encoding direction of the complex image semantic vector can be controlled by the at least one intermediate data semantic vector, so as to obtain a complex image enhanced semantic vector through depth encoding. It should be noted that the original echo semantic vector carries original features characterizing the degree of interference. For example, high interference can be captured through deeper encoding to facilitate interference identification during decoding; low interference can be reduced through shallower encoding to decrease computational overhead. Therefore, the encoding depth of the complex image semantic vector can be controlled by the original echo semantic vector, thus balancing the accuracy of interference identification with computational overhead. Furthermore, since the intermediate data semantic vector is closer to the complex image semantic vector in the semantic space than the original echo semantic vector, controlling the encoding direction of the complex image semantic vector through the at least one intermediate data semantic vector can achieve higher accuracy than controlling the encoding direction through the original echo semantic vector.

[0050] It is understood that the specific method of semantic encoding of the original echo data in step S122 is not limited. For example, in an alternative implementation, in order to improve the accuracy of semantic encoding and make the formed original echo semantic vector carry more complex semantic information, step S122 may further include steps S122a, S122b, S122c and S122d, wherein the specific contents of each step are as follows.

[0051] Step S122a: Perform convolution processing on the original echo data to form the original echo convolution vector.

[0052] In this embodiment, the original echo data can be convolved to form an original echo convolution vector. It should be noted that the convolution process can be implemented using a convolution unit including a kernel (and may also include an activation function, etc.).

[0053] Step S122b: The original echo convolution vector is pooled and compressed to form an original echo pooling vector, and the original echo convolution vector is segmented to form multiple original echo segmentation vectors.

[0054] In this embodiment, after obtaining the original echo convolution vector, the original echo convolution vector can be pooled and compressed to form an original echo pooling vector, and then segmented to form multiple original echo segmentation vectors. That is, on the one hand, pooling compression can achieve size control (the size can be the same as the size of the original echo segmentation vector) while capturing important semantic information. On the other hand, segmentation can form multiple original echo segmentation vectors, each representing different local semantic information, while the original echo pooling vector represents global semantic information.

[0055] Step S122c: Focus the original echo pooling vector based on each of the original echo segmentation vectors to form a corresponding original echo focusing vector.

[0056] In this embodiment, after obtaining the original echo segmentation vector and the original echo pooling vector, the original echo pooling vector can be focused based on each original echo segmentation vector to form a corresponding original echo focusing vector. That is, global semantic information can be focused based on local semantic information, resulting in focusing based on each original echo segmentation vector with small granularity, focusing only on the correlation between a portion of the global semantic information, thus improving focusing accuracy. The focusing method can be either a gating mechanism or an attention mechanism; no specific limitation is made here. For example, since the semantic information is of the same dimension and has a similar semantic space, focusing can be based on a gating mechanism to reduce the computational overhead of focusing.

[0057] Step S122d: Aggregate each of the original echo focusing vectors to obtain the original echo semantic vector.

[0058] In this embodiment of the application, after obtaining the original echo focusing vector, each of the original echo focusing vectors can be aggregated to obtain the original echo semantic vector. For example, the original echo focusing vectors can be concatenated to obtain the echo semantic vector, thus achieving a balance between semantic accuracy and semantic richness.

[0059] It is understood that in step S123 above, the specific method of semantic encoding of the data after processing at least one node before the third node in the target processing is not limited. For example, in an alternative implementation, in order to improve the accuracy of semantic encoding and make the resulting intermediate data semantic vector carry more complex semantic information, step S123 above may further include steps S123a, S123b, S123c and S123d, wherein the specific contents of each step are as follows.

[0060] Step S123a: Perform convolution processing on the azimuth compressed data obtained by performing azimuth compression processing on the second node to form an azimuth compressed convolution vector.

[0061] In this embodiment, the azimuth-compressed data obtained by performing azimuth compression processing on the second node can be convolved to form an azimuth-compressed convolution vector. It should be noted that the convolution processing can be implemented using a convolution unit including a convolution kernel (and may also include an activation function, etc.).

[0062] Step S123b: Perform pooling processing on the azimuth compressed convolution vector to form an azimuth compressed pooling vector, and segment the azimuth compressed convolution vector to form multiple azimuth compressed segmentation vectors.

[0063] In this embodiment, after obtaining the azimuth compressed convolution vector, the azimuth compressed convolution vector can be pooled to form an azimuth compressed pooling vector, and then segmented to form multiple azimuth compressed segmentation vectors. That is, on the one hand, pooling compression can achieve size control (the size can be the same as the size of the azimuth compressed segmentation vector) while capturing important semantic information. On the other hand, segmentation can form multiple azimuth compressed segmentation vectors, each representing different local semantic information, while the azimuth compressed pooling vector represents global semantic information.

[0064] Step S123c: Focus the azimuth compression pooling vector based on each of the azimuth compression segmentation vectors to form a corresponding azimuth compression focusing vector.

[0065] In this embodiment, after obtaining the azimuth compression segmentation vector and the azimuth compression pooling vector, the azimuth compression pooling vector can be focused based on each azimuth compression segmentation vector to form a corresponding azimuth compression focus vector. That is, global semantic information can be focused based on local semantic information, so that the focusing based on each azimuth compression segmentation vector has a small granularity, focusing on the correlation between only a portion of the global semantic information, thus improving the focusing accuracy. The focusing method can be either a gating mechanism or an attention mechanism; no specific limitation is made here. For example, since the semantic information is of the same dimension and the semantic space is similar, focusing can be based on a gating mechanism to reduce the computational overhead of focusing.

[0066] Step S123d: Aggregate each of the azimuth compression focusing vectors to obtain the intermediate data semantic vector.

[0067] In this embodiment, after obtaining the azimuth compression and focusing vector, each of the azimuth compression and focusing vectors can be aggregated to obtain an intermediate data semantic vector. For example, the azimuth compression and focusing vectors can be concatenated to obtain the intermediate data semantic vector, thus achieving a balance between semantic accuracy and semantic richness.

[0068] It is understood that the specific method of semantic encoding of the original SAR complex image in step S124 above is not limited. For example, in an alternative implementation, in order to improve the accuracy of semantic encoding and make the semantic vector of the formed complex image carry more complex semantic information, step S124 above may further include steps S124a, S124b, S124c and S124d, wherein the specific contents of each step are as follows.

[0069] Step S124a: Perform convolution processing on the original SAR complex image to form a complex image convolution vector.

[0070] In this embodiment, the original SAR complex image can be convolved to form a complex image convolution vector. It should be noted that the convolution process can be implemented using a convolution unit including a kernel (and may also include an activation function, etc.).

[0071] Step S124b: Perform pooling processing on the complex image convolution vector to form a complex image pooling vector, and segment the complex image convolution vector to form multiple complex image segmentation vectors.

[0072] In this embodiment, after obtaining the complex image convolution vector, the complex image convolution vector can be pooled to form a complex image pooling vector, and then segmented to form multiple complex image segmentation vectors. That is, on the one hand, pooling compression can be used to capture important semantic information while achieving size control (the size can be the same as the size of the complex image segmentation vector). On the other hand, segmentation can be performed to form multiple complex image segmentation vectors, each representing different local semantic information, while the complex image pooling vector represents global semantic information.

[0073] Step S124c: Focus the complex image pooling vector based on each of the complex image segmentation vectors to form a corresponding complex image focusing vector.

[0074] In this embodiment, after obtaining the complex image segmentation vector and the complex image pooling vector, the complex image pooling vector can be focused based on each of the complex image segmentation vectors to form corresponding complex image focus vectors. That is, global semantic information can be focused based on local semantic information, so that the focusing based on each of the complex image segmentation vectors has a small granularity, focusing only on the correlation between a portion of the global semantic information, resulting in higher focusing accuracy. The focusing method can be either a gating mechanism or an attention mechanism; no specific limitation is made here. For example, since the semantic information is of the same dimension and the semantic space is similar, focusing can be based on a gating mechanism to reduce the computational overhead of focusing.

[0075] Step S124d: Aggregate each of the complex image focus vectors to obtain the complex image semantic vector.

[0076] In this embodiment, after obtaining the complex image focus vectors, each complex image focus vector can be aggregated to obtain a complex image semantic vector. For example, the complex image focus vectors can be concatenated to obtain the echo semantic vector, thus achieving a balance between semantic accuracy and semantic richness.

[0077] It is understood that in step S125 above, the specific method of obtaining the complex image enhanced semantic vector through depth coding is not limited. For example, in an alternative implementation, in order to effectively control the coding depth so that depth coding can effectively balance coding accuracy and coding overhead, step S125 above may further include steps S125a and S125b, wherein the specific contents of each step are as follows.

[0078] Step S125a: Based on the embedding vectors corresponding to the target signal power, noise floor power, and signal-to-noise ratio in the original echo data, the original echo semantic vector is focused and mined to form an interference degree characterization vector. Based on the interference degree characterization vector, a mapping output is performed to obtain the target depth.

[0079] In this embodiment, based on the embedding vectors corresponding to the target signal power, noise floor power, and signal-to-noise ratio (SNR) in the original echo data (which can be obtained by word embedding of the target signal power, noise floor power, and SNR using a trained word embedding model; the target signal power, noise floor power, and SNR can be estimated; specific estimation schemes can refer to relevant existing technologies), the original echo semantic vector is focused and mined (this can be implemented based on a gating mechanism or an attention mechanism) to form an interference level representation vector. Then, based on the interference level representation vector, a mapping output is performed (for example, by performing fully connected processing on the interference level standard vector, and then activating the output to obtain a parameter used to represent the target depth), to obtain the target depth. The target depth is used to indicate the encoding depth of the complex image semantic vector. It should be noted that although the original echo semantic vector is formed by encoding the original echo data and carries semantics related to the degree of noise interference, it is not very specific. Therefore, the semantic information represented by the embedding vector corresponding to the target signal power, noise floor power, and signal-to-noise ratio can be focused and mined to make the resulting interference degree representation vector reliably represent the degree of interference.

[0080] Step S125b: According to the encoding direction represented by the at least one intermediate data semantic vector, perform depth encoding on the complex image semantic vector corresponding to the target depth to form a complex image enhanced semantic vector.

[0081] In this embodiment of the application, after obtaining the target depth, the complex image semantic vector can be depth encoded according to the encoding direction represented by the at least one intermediate data semantic vector to form a complex image enhanced semantic vector.

[0082] It is understood that in step S125b above, the specific method of performing depth encoding on the complex image semantic vector corresponding to the target depth is not limited. For example, in an alternative implementation, in order to ensure the reliability of depth encoding and make the semantic representation accuracy of the resulting complex image enhanced semantic vector higher, step S125b above may further include steps b1, b2, b3 and b4, wherein the specific contents of each step are as follows.

[0083] Step b1: Perform semantic space transformation on the at least one intermediate data semantic vector to form a transformed data semantic vector.

[0084] In the embodiments of this application, combined with Figure 4 The semantic space transformation can be performed on the at least one intermediate data semantic vector to form a transformed data semantic vector. For example, semantic space transformation can be achieved through linear mapping. This not only captures linear relationships but also transforms intermediate data semantic vectors belonging to different semantic spaces into the semantic space of the complex image semantic vector, thereby ensuring the accuracy of subsequent focusing. Furthermore, it should be noted that when there are multiple intermediate data semantic vectors, semantic space transformation can be performed separately for each vector, and the results of the semantic space transformation can be processed by averaging, etc., to obtain the transformed data semantic vector.

[0085] Step b2: In the depth encoding of the first depth, the complex image semantic vector is weighted and residually connected based on the focus direction representation parameter of the transformed data semantic vector or the focus direction representation parameter of the related semantic representation between the transformed data semantic vector and the complex image semantic vector to obtain the complex image depth semantic vector of the first depth.

[0086] In this embodiment, after obtaining the transformed data semantic vector, in the depth encoding of the first depth, the complex image semantic vector can be weighted based on the focus direction representation parameter represented by the transformed data semantic vector (for example, the transformed data semantic vector can be gated to obtain the corresponding gate parameter distribution, which serves as the focus direction representation parameter) or based on the focus direction representation parameter of the related semantic representation between the transformed data semantic vector and the complex image semantic vector (for example, the attention parameter between the transformed data semantic vector and the complex image semantic vector can be calculated, which serves as the focus direction representation parameter). (For example, when performing gated mapping to obtain the focus direction representation parameter, the focus direction representation parameter and the complex image semantic vector can be multiplied bitwise; when performing attention processing to obtain the focus direction representation parameter, the complex image semantic vector can be weighted and summed based on the focus direction representation parameter). A residual connection is then performed (i.e., the weighted result is added to the complex image semantic vector) to obtain the complex image depth semantic vector for the first depth.

[0087] Step b3: In each depth encoding between the second depth and the target depth, the complex image depth semantic vector of the previous depth is weighted and residually connected based on the focus direction representation parameter of the focus vector of the transformed data semantic vector at the corresponding depth or the focus direction representation parameter of the related semantic representation between the focus vector of the transformed data semantic vector at the corresponding depth and the complex image depth semantic vector of the previous depth, so as to obtain the complex image depth semantic vector of the corresponding depth.

[0088] In this embodiment, in each depth encoding between the second depth and the target depth, the complex image depth semantic vector of the previous depth is weighted and residually connected based on the focus direction representation parameter represented by the focus vector of the transformed data semantic vector at the corresponding depth, or based on the focus direction representation parameter of the related semantic representation between the focus vector of the transformed data semantic vector at the corresponding depth and the complex image depth semantic vector of the previous depth, to obtain the complex image depth semantic vector of the corresponding depth. For example, in the depth encoding of the second depth, the complex image depth semantic vector of the first depth is weighted and residually connected based on the focus direction representation parameter represented by the focus vector of the transformed data semantic vector at the second depth (such as the vector obtained by self-attention processing of the transformed data semantic vector), or based on the focus direction representation parameter of the related semantic representation between the focus vector of the transformed data semantic vector at the second depth and the complex image depth semantic vector of the first depth, to obtain the complex image depth semantic vector of the second depth. For example, in the depth encoding of the third depth, the complex image depth semantic vector of the second depth is weighted and residually connected based on the focus direction representation parameter of the focus vector of the transformed data semantic vector in the third depth (such as the vector obtained by performing self-attention processing on the focus vector of the transformed data semantic vector in the second depth) or the focus direction representation parameter of the related semantic representation between the focus vector of the transformed data semantic vector in the third depth and the complex image depth semantic vector of the second depth, so as to obtain the complex image depth semantic vector of the third depth.

[0089] Step b4: Determine the complex image depth semantic vector of the target depth as the complex image enhancement semantic vector.

[0090] In this embodiment of the application, after obtaining the complex image depth semantic vector of the target depth, the complex image depth semantic vector of the target depth can be determined as the complex image enhancement semantic vector.

[0091] Thirdly, regarding step S130, it should be noted that the specific method for decoding the complex image enhancement semantic vector is not limited and can be selected according to actual needs. For example, in an alternative implementation, a fully connected mapping can be performed on the complex image enhancement semantic vector, and then a linear mapping or identity mapping can be performed on the obtained fully connected mapping vector to obtain the corresponding interference probability distribution map. The size of the interference probability distribution map is the same as the size of the original SAR complex image, and each probability in the interference probability distribution map represents the probability that a pixel at a corresponding position in the original SAR complex image belongs to an interference region. Then, each probability in the interference probability distribution map can be compared with a pre-set threshold (such as 0.7), and the regions corresponding to pixels with probabilities greater than the threshold are determined as interference regions.

[0092] Fourthly, regarding step S140, it should be noted that the specific method for suppressing interference in the deception interference region of the original SAR complex image is not limited and can be selected according to actual needs. For example, in an alternative implementation, the pixel value of the pixel belonging to the deception interference region can be updated to 0, or the pixel value of the pixel belonging to the deception interference region can be attenuated according to a certain ratio to achieve interference suppression.

[0093] Fifthly, regarding the SAR anti-spoofing and jamming method based on deep learning networks, it is necessary to further explain that it may also include a step of training to form the anti-spoofing and jamming identification network. In an alternative implementation, this step may include the following: First, the original echo data of the sample can be processed to form the original SAR complex image of the sample. Please refer to the relevant explanation of step S110 above. Secondly, using the coding model in the initial anti-spoofing interference identification network, based on the semantic information in the original echo data of the sample and the data processed by at least one node before the third node in the target processing, the semantic information of the original SAR complex image of the sample is encoded and enhanced to form a sample complex image enhanced semantic vector. For the relevant explanation of step S120 above, please refer to the previous text. Then, using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the sample complex image is decoded to obtain the corresponding spoofing interference region recognition result. Refer to the relevant explanation of step S130 above. Finally, based on the error between the deception interference region identification result corresponding to the original echo data of the sample and the deception interference region label corresponding to the original echo data of the sample (the error calculation method is not limited and can be selected according to actual needs), the anti-deception interference identification network is trained (e.g., the network parameters of the anti-deception interference identification network are updated and adjusted along the direction of reducing error until the error converges), and the trained anti-deception interference identification network is obtained.

[0094] Combination Figure 5 This application also provides a SAR anti-spoofing jamming device based on a deep learning network that can be applied to the aforementioned electronic devices. The SAR anti-spoofing jamming device based on a deep learning network may include an echo data processing module, a data encoding module, a semantic decoding module, and an interference suppression module.

[0095] The echo data processing module is used to perform target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image. The target processing includes at least sequential range compression processing at a first node, azimuth compression processing at a second node, and complex image generation processing at a third node. In this embodiment, the echo data processing module can be used to perform... Figure 2 The relevant content regarding the echo data processing module in step S110 shown can be found in the previous description of step S110.

[0096] The data encoding module is used to encode and enhance the semantic information of the original SAR complex image using an encoding model in a deep learning-based anti-spoofing interference recognition network, based on the semantic information in the original echo data and the processed data of at least one node before the third node in the target processing, to form a complex image enhanced semantic vector. In this embodiment, the data encoding module can be used to perform... Figure 2 The relevant content regarding the data encoding module in step S120 shown can be found in the preceding description of step S120.

[0097] The semantic decoding module is used to decode the enhanced semantic vector of the complex image using the decoding model in the anti-spoofing interference recognition network to obtain the spoofing interference region recognition result. In this embodiment, the semantic decoding module can be used to perform... Figure 2 The relevant content regarding the semantic decoding module in step S130 shown can be found in the preceding description of step S130.

[0098] The interference suppression module is used to suppress the deception interference regions in the original SAR complex image based on the deception interference region identification result, thereby obtaining an anti-deception interference SAR complex image. In this embodiment, the interference suppression module can be used to perform... Figure 2 The relevant content regarding the interference suppression module in step S140 shown can be found in the previous description of step S140.

[0099] In summary, the SAR anti-spoofing interference method, apparatus, and device based on deep learning networks provided in this application firstly performs target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image; secondly, based on the semantic information in the raw echo data and the data processed by at least one node before the third node in the target processing, the semantic information of the raw SAR complex image is encoded and enhanced to form a complex image enhanced semantic vector; then, the complex image enhanced semantic vector is decoded to obtain the deception interference region identification result; finally, based on the deception interference region identification result, interference suppression is performed on the deception interference region in the raw SAR complex image to obtain an anti-spoofing interference SAR complex image. Based on the above, since the encoding process not only mines the semantic information in the formed raw SAR complex image but also enhances the semantics based on the raw echo data and the intermediate processed data, the semantic representation accuracy of the formed complex image enhanced semantic vector is higher. Therefore, the reliability of the identification result obtained from decoding is higher, thereby ensuring the reliability of interference suppression. Thus, the solution provided in this application can improve the problem of relatively low reliability in anti-spoofing interference in existing technologies.

[0100] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0101] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0102] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A SAR anti-spoofing interference method based on deep learning networks, characterized in that, include: The raw echo data received by the SAR receiver is subjected to target processing to form a raw SAR complex image. The target processing includes at least the range compression processing of the first node, the azimuth compression processing of the second node, and the complex image generation processing of the third node performed sequentially. The original echo data, the data processed by at least one node before the third node in the target processing, and the original SAR complex image are respectively loaded into the coding model of the anti-spoofing interference recognition network based on deep learning; the original echo data is semantically encoded to form an original echo semantic vector; the data processed by at least one node before the third node in the target processing is semantically encoded to form at least one corresponding intermediate data semantic vector; the original SAR complex image is semantically encoded to form a complex image semantic vector; in the deep coding process of the complex image semantic vector, the coding depth of the complex image semantic vector is controlled by the original echo semantic vector, and the coding direction of the complex image semantic vector is controlled by the at least one intermediate data semantic vector, so as to obtain a complex image enhanced semantic vector through deep coding; Using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the complex image is decoded to obtain the identification result of the spoofing interference region; Based on the identification results of the deception interference region, interference suppression is performed on the deception interference region in the original SAR complex image to obtain an anti-deception interference SAR complex image.

2. The SAR anti-spoofing interference method based on deep learning networks according to claim 1, characterized in that, The step of controlling the encoding depth of the complex image semantic vector through the original echo semantic vector and controlling the encoding direction of the complex image semantic vector through at least one intermediate data semantic vector, in order to obtain the complex image enhanced semantic vector through depth encoding, includes: Based on the embedding vectors corresponding to the target signal power, noise floor power, and signal-to-noise ratio in the original echo data, the original echo semantic vector is focused and mined to form an interference level characterization vector. Based on the interference level characterization vector, a mapping output is performed to obtain the target depth, wherein the target depth is used to indicate the encoding depth of the complex image semantic vector. According to the encoding direction represented by the at least one intermediate data semantic vector, the complex image semantic vector is depth encoded corresponding to the target depth to form a complex image enhanced semantic vector.

3. The SAR anti-spoofing interference method based on deep learning networks according to claim 2, characterized in that, The step of performing depth encoding on the complex image semantic vector according to the encoding direction represented by the at least one intermediate data semantic vector to form a complex image enhanced semantic vector includes: Perform semantic space transformation on the at least one intermediate data semantic vector to form a transformed data semantic vector; In the depth encoding of the first depth, the complex image semantic vector is weighted and residually connected based on the focus direction representation parameter of the transformed data semantic vector or the focus direction representation parameter of the related semantic representation between the transformed data semantic vector and the complex image semantic vector to obtain the complex image depth semantic vector of the first depth. In each depth encoding between the second depth and the target depth, the complex image depth semantic vector of the previous depth is weighted and residually connected based on the focus direction representation parameter represented by the focus vector of the transformed data semantic vector at the corresponding depth or the focus direction representation parameter based on the relevant semantic representation between the focus vector of the transformed data semantic vector at the corresponding depth and the complex image depth semantic vector of the previous depth, so as to obtain the complex image depth semantic vector of the corresponding depth. The depth semantic vector of the complex image at the target depth is determined as the complex image enhancement semantic vector.

4. The SAR anti-spoofing interference method based on deep learning networks according to claim 1, characterized in that, The step of semantically encoding the original echo data to form an original echo semantic vector includes: The original echo data is convolved to form the original echo convolution vector; The original echo convolution vector is pooled and compressed to form an original echo pooling vector, and the original echo convolution vector is segmented to form multiple original echo segmentation vectors. Based on each of the original echo segmentation vectors, the original echo pooling vectors are focused to form each corresponding original echo focusing vector; By aggregating each of the original echo focusing vectors, the original echo semantic vector is obtained.

5. The SAR anti-spoofing interference method based on deep learning networks according to claim 1, characterized in that, The step of semantically encoding the data processed by at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector includes: The azimuth-compressed data obtained by the second node is convolved to form an azimuth-compressed convolution vector. The azimuth compressed convolution vector is pooled to form an azimuth compressed pooling vector, and the azimuth compressed convolution vector is segmented to form multiple azimuth compressed segmentation vectors. Based on each of the azimuth compression segmentation vectors, the azimuth compression pooling vectors are focused to form corresponding azimuth compression focusing vectors; By aggregating each of the aforementioned orientation compression focus vectors, an intermediate data semantic vector is obtained.

6. The SAR anti-spoofing interference method based on deep learning networks according to claim 1, characterized in that, The step of semantically encoding the original SAR complex image to form a complex image semantic vector includes: The original SAR complex image is convolved to form a complex image convolution vector; The complex image convolution vector is pooled to form a complex image pooling vector, and the complex image convolution vector is segmented to form multiple complex image segmentation vectors; Based on each of the complex image segmentation vectors, the complex image pooling vectors are focused to form a corresponding complex image focusing vector; By aggregating the focus vectors of each complex image, a complex image semantic vector is obtained.

7. The SAR anti-spoofing interference method based on deep learning networks according to any one of claims 1-6, characterized in that, Also includes: The original echo data of the sample is processed to form the original SAR complex image of the sample. Using the coding model in the initial anti-spoofing interference identification network, based on the semantic information in the original echo data of the sample and the data processed by at least one node before the third node in the target processing, the semantic information of the original SAR complex image of the sample is encoded and enhanced to form a sample complex image enhanced semantic vector. Using the decoding model in the anti-spoofing interference recognition network, the enhanced semantic vector of the sample complex image is decoded to obtain the corresponding spoofing interference region recognition result; Based on the error between the spoofing interference region identification result corresponding to the original echo data of the sample and the spoofing interference region label corresponding to the original echo data of the sample, the anti-spoofing interference identification network is trained to obtain the trained anti-spoofing interference identification network.

8. A SAR anti-spoofing jamming device based on deep learning networks, characterized in that, include: The echo data processing module is used to perform target processing on the raw echo data received by the SAR receiver to form a raw SAR complex image. The target processing includes at least the range compression processing of the first node, the azimuth compression processing of the second node, and the complex image generation processing of the third node performed sequentially. The data encoding module is used to load the original echo data, the data processed by at least one node before the third node in the target processing, and the original SAR complex image into the encoding model of the anti-spoofing interference recognition network based on deep learning; to perform semantic encoding on the original echo data to form an original echo semantic vector; to perform semantic encoding on the data processed by at least one node before the third node in the target processing to form at least one corresponding intermediate data semantic vector; to perform semantic encoding on the original SAR complex image to form a complex image semantic vector; and during the depth encoding process of the complex image semantic vector, to control the encoding depth of the complex image semantic vector through the original echo semantic vector, and to control the encoding direction of the complex image semantic vector through the at least one intermediate data semantic vector, so as to obtain a complex image enhanced semantic vector through depth encoding. The semantic decoding module is used to decode the enhanced semantic vector of the complex image using the decoding model in the anti-spoofing interference recognition network to obtain the identification result of the spoofing interference region; The interference suppression module is used to suppress the deception interference region in the original SAR complex image based on the deception interference region identification result, so as to obtain an anti-deception interference SAR complex image.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the SAR anti-spoofing jamming method based on deep learning networks as described in any one of claims 1-7.

Images