An electronic reconnaissance system and method

Abstract

The present application discloses an electronic reconnaissance system and method. The system includes a radio frequency module, a feature extraction module, and a signal processing module. The radio frequency module collects signals emitted by radiation sources in multiple channels; the feature extraction module extracts features from each collected signal to obtain the global features and local features of each signal; the signal processing module includes signal processing channels corresponding to the number of channels of the collected signals, and any one of the signal processing channels includes a first cross-attention layer, a feature splicing layer, a first normalization layer, a fully connected layer, a second normalization layer, a second cross-attention layer, and a second feature splicing layer. The signal processing channels process the signal feature information to obtain a third spliced feature, and the signal processing module performs signal detection, identification, and direction finding of the radiation source based on the third spliced feature. The present application can achieve efficient and accurate processing of radiation source signals in a complex communication environment, and quickly and accurately complete the tasks of signal detection, identification, and direction finding of the radiation source.

Description

Technical Field

[0001] This application relates to the field of electronic detection technology, and particularly to an electronic reconnaissance system and method. Background Art

[0002] In the fields of modern communication and electronic countermeasures, the signal detection, identification, and direction finding technologies of radiation sources have extremely important application values. Signal detection is a basic task in the fields of communication and electronic reconnaissance, and is responsible for capturing the signs of the existence of radiation source signals from complex electromagnetic environments. Signal identification is to deeply analyze signal characteristics and accurately classify signals based on detection. The two provide key support for the preliminary screening and accurate classification of signals in complex electromagnetic environments. Radiation source direction finding plays an indispensable role in scenarios such as electronic reconnaissance, radio management, and non-cooperative communication. For example, in electronic reconnaissance, by direction finding and identifying radiation sources, technical parameters, geographical locations, and threat levels of enemy radars, communication devices, etc. can be obtained, providing key information for tactical decisions.

[0003] In the prior art, most can only achieve one of the tasks of detection, identification, or direction finding, or combine two of them. This is mainly because multi-source signal interference, large fluctuations in signal intensity, and high complexity of signal frequencies make it difficult for a single processing process to simultaneously take into account the accuracy of signal detection, identification, and direction finding. Summary of the Invention

[0004] To solve the deficiencies of the prior art, the following technical solutions are adopted in this application:

[0005] In a first aspect, an electronic reconnaissance system provided by this application, the electronic reconnaissance system includes:

[0006] An RF module configured to collect signals emitted by a radiation source in multiple channels;

[0007] A feature extraction module configured to extract features from each collected signal to obtain the global features and local features of each signal respectively;

[0008] A signal processing module includes signal processing channels corresponding to the number of channels of the collected signals, and any one of the signal processing channels includes:

[0009] A first cross-attention layer including two multi-head attention units with shared weights, where the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features, and the query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features;

[0010] The feature splicing layer includes two feature splicing units, and the feature splicing units are configured to splice and integrate the outputs of the first multi-head attention unit and the second multi-head attention unit respectively to obtain a first spliced feature and a second spliced feature;

[0011] The first normalization layer includes a first normalization unit and a second normalization unit, and is configured to perform layer normalization processing on the outputs of the two feature splicing units respectively;

[0012] The second normalization layer includes a third normalization unit and a fourth normalization unit. The input end of the third normalization unit is connected to the output end of the first normalization unit through a fully connected layer, and is connected to the output end of the first normalization unit in a residual connection manner; the input end of the fourth normalization unit is connected to the output end of the second normalization unit through a fully connected layer, and is connected to the output end of the second normalization unit in a residual connection manner;

[0013] The second cross-attention layer includes a third multi-head attention unit and a fourth multi-head attention unit with shared weights. The query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the outputs of the third normalization unit. The query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the outputs of the fourth normalization unit;

[0014] The second feature splicing layer splices and integrates the output of the second cross-attention layer to obtain a third spliced feature, and the signal processing module performs signal detection, identification, and direction finding of the radiation source based on the third spliced feature.

[0015] In summary, an electronic reconnaissance system provided by the present application adopts multiple paths in parallel to extract features from the collected signals, efficiently extracts the global features and local features of the radiation source signals, and adopts a processing architecture with multiple signal processing channels in parallel. Each channel corresponds to an independent processing flow of one path of signals, realizing the efficient parallel processing of multiple paths of signals. Combining methods such as the multi-head attention mechanism, shared weight design, residual connection, and layer normalization, it realizes the accurate analysis and complete fusion of signal features, so as to realize the efficient and accurate processing of radiation source signals in the case of multi-source signal interference, large signal intensity changes, and complex signal frequencies, and quickly and accurately complete the tasks of signal detection, identification, and direction finding of the radiation source.

[0016] Further, the system includes four of the signal processing channels, and the signal processing module further includes a phase difference calculation unit, and the phase difference calculation unit is configured to calculate the phase difference between the third spliced features of different signal processing channels to obtain a radiation source direction finding result.

[0017] Further, the system includes four of the signal processing channels, and the signal processing module further includes a first fully connected layer, which is configured to compare the third concatenated features of the four signal processing channels with a preset signal presence determination criterion respectively, and output a determination result on whether a signal exists.

[0018] Further, the system includes four of the signal processing channels, and the signal processing module further includes a second fully connected layer, which is configured to match and classify the third concatenated features of the four signal processing channels with the features of different signal types that have been learned respectively, and output the modulation type of the signal.

[0019] Further, the feature extraction module includes:

[0020] A Hilbert transform unit, which is configured to perform a Hilbert transform on each acquired signal to obtain a Hilbert transform signal;

[0021] A first integration unit, which is configured to concatenate the Hilbert transform signal and the original acquired signal to obtain an integrated signal;

[0022] A first long short-term memory network, which is configured to perform feature extraction on the integrated signal to obtain the global feature.

[0023] Further, the feature extraction module further includes:

[0024] A sliding window cutting unit, which is configured to divide the integrated signal into a plurality of signal segments;

[0025] A second long short-term memory network, which is configured to perform feature extraction on each of the signal segments;

[0026] A second integration unit, which is configured to perform signal concatenation on the features extracted by the second long short-term memory network to obtain the local feature.

[0027] In a second aspect, the present application further provides an electronic reconnaissance method, and the method includes:

[0028] Collect signals emitted by a radiation source through multiple channels;

[0029] Perform feature extraction on each acquired signal to obtain the global feature and the local feature of each signal respectively;

[0030] Each signal is processed through multi-channel parallel processing. Any signal processing channel processes the global features and local features through two weight-sharing multi-head attention units. Among them, the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features. The query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features;

[0031] The output of the first multi-head attention unit and the second multi-head attention unit are respectively spliced and integrated through a feature splicing unit to obtain a first spliced feature and a second spliced feature;

[0032] Through a first normalization unit and a second normalization unit, layer normalization processing is respectively performed on the outputs of the two feature splicing units;

[0033] The output of the first normalization unit is transmitted to a third normalization unit respectively through a fully connected layer and a residual connection, and the output of the second normalization unit is transmitted to a fourth normalization unit. The third normalization unit and the fourth normalization unit respectively perform layer normalization processing on the received signals;

[0034] The outputs of the third normalization unit and the fourth normalization unit are weighted and fused through a weight-sharing third multi-head attention unit and a fourth multi-head attention unit. Among them, the query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the outputs of the third normalization unit. The query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the outputs of the fourth normalization unit;

[0035] The output results of the third multi-head attention unit and the fourth multi-head attention unit are spliced and integrated to obtain a third spliced feature, and signal detection, recognition, and direction finding of the radiation source are performed based on the third spliced feature.

[0036] Further, four signal processing channels for parallel processing of signals are set, and the phase difference between the third spliced features of different signal processing channels is calculated to obtain the radiation source direction finding result.

[0037] Further, Hilbert transform is performed on each collected signal to obtain a Hilbert transform signal;

[0038] The Hilbert transform signal is spliced with the original collected signal to obtain an integrated signal;

[0039] A long short-term memory network is used to extract features from the integrated signal to obtain the global features.

[0040] Further, the integrated signal is segmented into a plurality of signal segments;

[0041] The long short-term memory network is used to extract features from each of the signal segments;

[0042] The features extracted by the long short-term memory network are signal-spliced to obtain the local features. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the composition modules of an electronic reconnaissance system provided by an embodiment of the present application;

[0044] Figure 2 It is a schematic diagram of the composition structure of a feature extraction module in an electronic reconnaissance system provided by an embodiment of the present application;

[0045] Figure 3 It is a schematic diagram of the composition structure of a first long short-term memory network in an electronic reconnaissance system provided by an embodiment of the present application;

[0046] Figure 4 It is a schematic diagram of the composition structure of another feature extraction module in an electronic reconnaissance system provided by an embodiment of the present application;

[0047] Figure 5 It is a schematic diagram of the composition structure of any signal processing channel in an electronic reconnaissance system provided by an embodiment of the present application;

[0048] Figure 6 It is a schematic diagram of the composition structure of a signal processing module in an electronic reconnaissance system provided by an embodiment of the present application;

[0049] Figure 7 It is a flowchart of the steps of an electronic reconnaissance method provided by an embodiment of the present application;

[0050] Figure 8 It is a flowchart of the steps of obtaining the global features of a signal in an electronic reconnaissance method provided by an embodiment of the present application;

[0051] Figure 9 It is a flowchart of the steps of obtaining the local features of a signal in an electronic reconnaissance method provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The present application will be described in detail below with reference to the specific embodiments shown in the drawings, but these embodiments do not limit the present application, and any structural, method, or functional transformation made by those of ordinary skill in the art based on these embodiments is included in the protection scope of the present application.

[0053] To solve the deficiencies of the prior art, in a first aspect, the present application provides an electronic reconnaissance system 100, asFigure 1 As shown in Figure 1 , the electronic reconnaissance system 100 includes: a radio frequency module 11, a feature extraction module 12, and a signal processing module 13. The radio frequency module 11 is configured to collect signals emitted by radiation sources in multiple channels; the feature extraction module 12 is configured to extract features from each collected signal to obtain the global features and local features of each signal respectively; the signal processing module 13 includes signal processing channels 131 corresponding to the number of channels of the collected signals, and any signal processing channel 131 includes: a first cross-attention layer 1311, a feature splicing layer 1312, a first normalization layer 1313, a second normalization layer 1314, a second cross-attention layer 1315, and a second feature splicing layer 1316.

[0054] The first cross-attention layer 1311 includes two weight-sharing multi-head attention units. Among them, the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features; the query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features; the feature splicing layer 1312 includes two feature splicing units, and the feature splicing units are configured to splice and integrate the outputs of the first multi-head attention unit and the second multi-head attention unit respectively to obtain the first spliced feature and the second spliced feature.

[0055] The first normalization layer 1313 includes a first normalization unit and a second normalization unit, and is configured to perform layer normalization processing on the outputs of the two feature splicing units respectively; the second normalization layer 1314 includes a third normalization unit and a fourth normalization unit. The input end of the third normalization unit is connected to the output end of the first normalization unit through a fully connected layer, and is connected to the output end of the first normalization unit in a residual connection manner; the input end of the fourth normalization unit is connected to the output end of the second normalization unit through a fully connected layer, and is connected to the output end of the second normalization unit in a residual connection manner;

[0056] The second cross-attention layer 1315 includes a third multi-head attention unit and a fourth multi-head attention unit with weight sharing. The query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the outputs of the third normalization unit; the query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the outputs of the fourth normalization unit; the second feature splicing layer 1316 splices and integrates the output of the second cross-attention layer 1315 to obtain the third spliced feature, and the signal processing module 13 performs signal detection, identification, and direction finding of the radiation source based on the third spliced feature.

[0057] Specifically, the radio frequency module 11 can be configured to have multiple antennas, and the signals emitted by the same radiation source are collected in multiple paths through the multiple antennas. The radio frequency module 11 is configured to perform preprocessing such as filtering, amplifying, and down-converting on the collected signals, and convert the collected signals into a form suitable for subsequent digital processing. The radio frequency module 11 is connected to the feature extraction module 12, and the radio frequency module 11 inputs the preprocessed collected signals into the feature extraction module 12 for the feature extraction module 12 to extract features from each path of the collected signals.

[0058] In the embodiment of the present application, the electronic reconnaissance system 100 includes multiple feature extraction modules 12, and the number of the feature extraction modules 12 is consistent with the number of signal paths collected by the radio frequency module 11. The feature extraction module 12 receives the multiple paths of signals collected by the radio frequency module 11, and each feature extraction module 12 separately extracts features from one path of the collected signals to obtain global features and local features that can characterize the characteristics of each path of signals. The global features reflect the overall behavior and change trend of the signals, such as the frequency change trend of the signals, the overall fluctuation of the power, etc.; the local features reflect the specific behavior and change trend of the signals within different time windows, such as the sudden change of the signals within a certain short time, the short-term fluctuation of a specific frequency resolution, etc. The global features and local features output by the feature extraction module 12 provide multi-dimensional data input for the cross-attention mechanism of the signal processing module 13 and provide rich data support for subsequent signal processing.

[0059] As an optional implementation manner, the feature extraction module 12 includes a Hilbert transform unit 121, a first integration unit 122, and a first long short-term memory network 123. The Hilbert transform unit 121 is configured to perform Hilbert transform on each path of the collected signals to obtain Hilbert transform signals; the first integration unit 122 is configured to splice the Hilbert transform signals and the original collected signals to obtain integrated signals; the first long short-term memory network 123 is configured to extract features from the integrated signals to obtain global features.

[0060] Specifically, as Figure 2 shown, each feature extraction module 12 includes a Hilbert transform unit 121, a first integration unit 122, and a first long short-term memory network 123. A feature extraction module 12 receives and processes one path of signals collected by the radio frequency module 11. The Hilbert transform unit 121 of the feature extraction module 12 first performs Hilbert transform on this path of signals, extracts the analytic signal from the real signal to obtain the Hilbert transform signal, and sends the Hilbert transform signal to the first integration unit 122. The Hilbert transform signal includes the instantaneous phase and amplitude information of the signal.

[0061] The first integration unit 122 is connected to the Hilbert transform unit 121. The first integration unit 122 receives the Hilbert transform signal and the original acquisition signal, splices the Hilbert transform signal and the original acquisition signal, integrates the time domain and phase information of the signals to obtain an integrated signal, and is connected to the first long short-term memory network 123. The first integration unit 122 sends the integrated signal to the first long short-term memory network 123. The first long short-term memory network 123 receives the integrated signal and extracts features from the integrated signal through the gating mechanism inside it, captures the dependencies and overall trends of the signal in the long time series, and obtains the global features of the acquisition signal.

[0062] Further, as an optional implementation, as Figure 3 shown, the first long short-term memory network 123 includes a forget gate, an input gate, and an output gate. Input the hidden state at the current time step x t , the hidden state at the previous time step h t-1 and the cell state at the previous time step G t-1 (the cell state is the memory unit, similar to a memory that can retain important information fragments for a long time). The forget gate outputs a value between 0 and 1 through the sigmoid function based on the hidden state at the previous time step h t-1 input and the current time step x t to determine which information to discard from the cell state.

[0063] Based on the hidden state at the previous time step h t-1 input and the current time step x t , the input gate determines which new information to write into the cell state through the sigmoid function. The cell state is updated according to the outputs of the forget gate and the input gate, that is, the output of the forget gate is multiplied by the cell state at the previous time step G t-1 , and the cell state discards some information. The output of the input gate is multiplied by the candidate cell state, and the candidate cell state adds some new information. After the two are added, the updated cell state is obtained, and thus the cell state at the current time step G t is output. The output gate outputs a value between 0 and 1 through the sigmoid function based on the hidden state at the previous time step h t-1 input and the current time step x t , and multiplies it with the current cell state processed by the tanh function Gt Multiply them to output the hidden state at the current time step h t The hidden state h t contains the global features of the acquired signal.

[0064] As an alternative implementation, the feature extraction module 12 further includes a sliding window cutting unit 124, a second long short-term memory network 125, and a second integration unit 126. The sliding window cutting unit 124 is configured to divide the integrated signal into multiple signal segments; the second long short-term memory network 125 is configured to extract features from each signal segment; the second integration unit 126 is configured to splice the features extracted by the second long short-term memory network 125 to obtain local features.

[0065] Specifically, as Figure 4 shown, each feature extraction module 12 further includes a sliding window cutting unit 124, a second long short-term memory network 125, and a second integration unit 126. The sliding window cutting unit 124 is connected to the first integration unit 122. The sliding window cutting unit 124 receives the integrated signal and divides the integrated signal into multiple signal segments each containing a certain time range. The sliding window cutting unit 124 is connected to multiple second long short-term memory networks 125. The number of second long short-term memory networks 125 can be the same as the number of signal segments. Each second long short-term memory network 125 independently processes one signal segment, and multiple second long short-term memory networks 125 process the signal segments in parallel to improve the signal processing speed. The second long short-term memory network 125 calculates based on the current signal segment and the state information of the previous time step. Its specific working mode is similar to that of the first long short-term memory network 123 described above and will not be elaborated here.

[0066] Different from the first long short-term memory network 123, the second long short-term memory network 125 processes signal segments within a short time range, and the second long short-term memory network 125 captures the dependency relationship and change characteristics of the signal within the local time window. The second integration unit 126 is connected to multiple second long short-term memory networks 125. The second long short-term memory network 125 sends the extracted signal features to the second integration unit 126, and the second integration unit 126 splices the received signal features to obtain local features that can reflect the specific behavior and changes of the signal within different time windows.

[0067] By performing feature extraction on each collected signal, the feature extraction module 12 respectively obtains the global features and local features of each signal. The feature extraction module 12 is connected to the signal processing module 13 and inputs the obtained signals of each path into the signal processing module 13. The signal processing module 13 includes multiple parallel signal processing channels 131. The number of signal processing channels 131 can correspond to the number of paths of the collected signals. Each signal processing channel 131 in the signal processing module 13 independently processes one signal to obtain the global features and local features of that signal. The multiple signal processing channels 131 process in parallel, realizing the efficient processing of multiple collected signals and shortening the signal processing time.

[0068] As an alternative implementation, as Figure 5 shown, any signal processing channel 131 includes a first cross-attention layer 1311, a feature splicing layer 1312, a first normalization layer 1313, a second normalization layer 1314, a second cross-attention layer 1315, and a second feature splicing layer 1316.

[0069] Exemplarily, the following takes the first signal processing channel 131 as an example for illustration, and takes the first signal processing channel 131 receiving and processing the global features and local features of the first collected signal as an example for illustration. It should be noted that in the electronic reconnaissance system 100 provided in this application, there is no restriction on which collected signal a signal processing channel 131 processes.

[0070] The first cross-attention layer 1311 of the first signal processing channel 131 is connected to a feature extraction module 12. The first cross-attention layer 1311 includes two first multi-head attention units MHA1 and second multi-head attention units MHA2 with shared weights. Among them, the first multi-head attention unit MHA1 uses the local feature as the query input, and uses the global feature as the key input and value input. The first multi-head attention unit MHA1 calculates the attention distribution of the local feature in the global feature space, such as the trend or correlation pattern of a certain signal segment in the entire time series. The second multi-head attention unit MHA2 uses the global feature as the query input, and uses the local feature as the key input and value input. The second multi-head attention unit MHA2 prompts the global feature to focus on the key segments in the local feature, such as sudden frequency offsets and instantaneous power changes in the signal. Through the multi-head attention mechanism, each attention head independently learns the feature associations of different information, enhancing the complementarity between the global feature and the local feature. Further, the design of weight sharing between the first multi-head attention unit MHA1 and the second multi-head attention unit MHA2 reduces the scale of calculation parameters while ensuring calculation accuracy, reduces the calculation complexity, and further improves the system processing efficiency.

[0071] The first cross-attention layer 1311 is connected to the feature concatenation layer 1312. The first cross-attention layer 1311 sends the processed feature information to the feature concatenation layer 1312. The feature concatenation layer 1312 includes a first feature concatenation unit CA1 and a second feature concatenation unit CA2. The first feature concatenation unit CA1 and the second feature concatenation unit CA2 are configured to perform dimensional concatenation on the feature information output by the first multi-head attention unit MHA1 and the second multi-head attention unit MHA2 respectively, integrate the interacted feature information, and form a first concatenated feature and a second concatenated feature, ensuring that subsequent processing can make decisions based on a complete feature set.

[0072] The feature concatenation layer 1312 is connected to the first normalization layer 1313. The feature concatenation layer 1312 sends the first concatenated feature and the second concatenated feature to the first normalization layer 1313. The first normalization layer 1313 includes a first normalization unit LN1 and a second normalization unit LN2. The first normalization unit LN1 and the second normalization unit LN2 are configured to perform normalization processing on the first concatenated feature and the second concatenated feature respectively. By performing normalization operations on the concatenated features, the scale difference problem that may exist in feature concatenation can be alleviated, which helps to stabilize the training process, accelerate the convergence speed, and improve the generalization ability of the system.

[0073] The first normalization layer 1313 is connected to the second normalization layer 1314. Specifically, the second normalization layer 1314 includes a third normalization unit LN3 and a fourth normalization unit LN4. Among them, the input end of the third normalization unit LN3 is connected to the output end of the first normalization unit LN1 through a fully connected layer, and the input end of the third normalization unit LN3 is also connected to the output end of the first normalization unit LN1 in a residual connection manner; the input end of the fourth normalization unit LN4 is connected to the output end of the second normalization unit LN2 through a fully connected layer, and the input end of the fourth normalization unit LN4 is also connected to the output end of the second normalization unit LN2 in a residual connection manner.

[0074] Exemplarily, taking the third normalization unit LN3 as an example for illustration, on the one hand, the third normalization unit LN3 receives the feature information output by the first normalization unit LN1 that has been mapped by the fully connected layer, and on the other hand, it receives the original output of the first normalization unit LN1 directly transmitted through the residual path. The fully connected layer can perform non-linear transformation and dimensional adjustment on the characteristic information output by the first normalization unit LN1 to capture more complex feature relationships; the residual connection allows the input information to be directly transmitted to the subsequent processing layer, avoiding the problem of gradient disappearance in the deep network, ensuring the integrity of the signal during the transmission process in the multi-layer network, and accelerating the training process to improve the system performance. The third normalization unit LN3 performs further layer normalization processing on the received feature information to further suppress the problem of covariate shift during the training process, prevent the computational model from falling into a local optimal solution during the training process, enable the computational model to better adapt to signals with different feature distributions, accelerate the convergence speed, and improve the generalization ability of the computational model.

[0075] The second normalization layer 1314 is connected to the second cross-attention layer 1315. Specifically, the second cross-attention layer 1315 includes a third multi-head attention unit MHA3 and a fourth multi-head attention unit MHA4 with shared weights. Among them, the third multi-head attention unit MHA3 uses the output of the fourth normalization unit LN4 as the query input, and the output of the third normalization unit LN3 as the key input and value input; the fourth multi-head attention unit MHA4 uses the output of the third normalization unit LN3 as the query input, and the output of the fourth normalization unit LN4 as the key input and value input. After calculating through the multi-head attention mechanism, this layer can capture the correlation information from different angles between the normalized signal features, such as the relationship between the signal phase difference and the modulation type, or the co-variation pattern of the signal features within different time windows, perform secondary interaction and fusion on the feature information, and thus output the result of weighted fusion of the input feature information.

[0076] The second cross-attention layer 1315 is connected to the second feature splicing layer 1316. The second feature splicing layer 1316 splices and integrates the two-way outputs of the second cross-attention layer 1315 along the feature dimension, integrates the multi-round interaction information of the global features and local features, and obtains the third spliced feature. The third spliced feature contains the signal phase feature, the modulation method of the feature signal, the coding feature, and the feature information related to the signal existence. The third spliced feature can reflect the attributes of the signal in different dimensions, such as the distribution feature of the signal strength, the frequency feature, etc.

[0077] Based on the above-obtained third spliced feature, the signal processing module 13 performs signal detection, recognition, and direction finding of the radiation source. For example, during the signal detection process, the information related to the signal existence in the third spliced feature is used to determine whether there is an effective signal, and thus whether there is a radiation source.

[0078] As an alternative implementation, as Figure 6 shown, the signal processing module 13 further includes a phase difference calculation unit 132. The phase difference calculation unit 132 is configured to be connected to all signal processing channels 131. The phase difference calculation unit 132 can obtain the third spliced features output by all signal processing channels 131. Among them, the phase information of each collected signal is included in the third spliced features output by each signal processing channel 131. The phase difference calculation unit 132 can calculate the difference in the phase information obtained by different signal processing channels 131, so as to calculate the direction of the radiation source and obtain the radiation source direction finding result.

[0079] As an alternative implementation, as Figure 6 shown, the signal processing module 13 further includes a plurality of first fully connected layers 133. The number of the first fully connected layers 133 can correspond to the number of signal processing channels 131. Each first fully connected layer 133 is respectively connected to a different signal processing channel 131. The first fully connected layer 133 is configured to obtain the third spliced features output by the connected signal processing channel 131. The first fully connected layer 133 compares the obtained third spliced features with a preset signal presence judgment criterion to judge whether there is a signal and output a judgment result. Based on the output results of all the first fully connected layers 133, the signal processing module 13 performs detection and comparison to obtain the detection result of the radiation source signal, thereby realizing the signal detection of the radiation source.

[0080] As an alternative implementation, as Figure 6 shown, the signal processing module 13 further includes a plurality of second fully connected layers 134. The number of the second fully connected layers 134 can correspond to the number of signal processing channels 131. Each second fully connected layer 134 is respectively connected to a different signal processing channel 131. The second fully connected layer 134 is configured to obtain the third spliced features output by the connected signal processing channel 131. The second fully connected layer 134 is configured to learn the features of different signal types, and match and classify the obtained third spliced features according to the learned features of different signal types, so as to determine and output category information such as the specific modulation method of the signal. Based on the output results of all the second fully connected layers 134, the signal processing module 13 performs identification and comparison to obtain the identification result of the radiation source signal, thereby realizing the signal identification of the radiation source. Further, when the modulation types and signal distributions of the multiple signals collected by the radio frequency module 11 are the same, the detection and identification accuracy of the radiation source signal can be further improved by comparing the detection and identification results of each signal.

[0081] As an alternative implementation, as Figure 6As shown in the figure, the electronic reconnaissance system 100 provided by the present application includes four signal processing channels 131, and a total of sixteen multi-head attention units sharing weights in pairs are adopted. Through parallel processing, each independent processing process corresponding to one path of signal is used to analyze the characteristics of multiple paths of signals simultaneously, ensuring the efficient processing of multiple paths of signals; and through the design of sharing weights, the number of model parameters is reduced, the computational complexity is reduced, and the processing efficiency of the system is further improved, so as to realize the fast processing of the collected signals in the case of multi-source signal interference, large signal strength variation, and complex signal frequency, etc., and be able to analyze and process the feature information in time after obtaining the signal features, and quickly complete the signal detection, identification, and direction finding tasks of the radiation source, so as to accurately judge the position, type, and threat level of the radiation source.

[0082] According to the above description, an electronic reconnaissance system 100 provided by the present application adopts a multi-stage feature extraction process combining Hilbert transform, sliding window cutting, and long short-term memory network, extracts the features of the collected signals in a multi-path parallel manner, efficiently extracts the global features and local features of the radiation source signals, and adopts a processing architecture with multiple signal processing channels 131 in parallel. Each channel corresponds to an independent processing process of one path of signal, realizing the efficient parallel processing of multiple paths of signals. Combining methods such as multi-head attention mechanism, shared weight design, residual connection, and layer normalization, etc., realizes the accurate analysis and complete fusion of signal features, so as to efficiently complete the all-round processing of the radiation source signals in a complex communication environment and quickly complete the signal detection, identification, and direction finding tasks of the radiation source.

[0083] In the second aspect, the present application also provides an electronic reconnaissance method, as Figure 7 shown, the method includes the following steps:

[0084] Step S11, collect signals emitted by the radiation source in multiple paths.

[0085] Step S12, extract the features of each path of the collected signals, and respectively obtain the global features and local features of each path of signals.

[0086] Step S13, process each path of signal in multi-channel parallel. Any signal processing channel 131 processes the global features and local features through two multi-head attention units sharing weights. Among them, the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features. The query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features.

[0087] Step S14, respectively splice and integrate the outputs of the first multi-head attention unit and the second multi-head attention unit through a feature splicing unit to obtain a first spliced feature and a second spliced feature.

[0088] Step S15: Through the first normalization unit and the second normalization unit, perform layer normalization processing on the outputs of the two feature splicing units respectively.

[0089] Step S16: Respectively, through the fully connected layer and the residual connection method, transfer the output of the first normalization unit to the third normalization unit, and transfer the output of the second normalization unit to the fourth normalization unit. The third normalization unit and the fourth normalization unit respectively perform layer normalization processing on the received signals.

[0090] Step S17: Through the third multi-head attention unit and the fourth multi-head attention unit with weight sharing, perform weighted fusion on the outputs of the third normalization unit and the fourth normalization unit. Among them, the query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the outputs of the third normalization unit. The query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the outputs of the fourth normalization unit.

[0091] Step S18: Concatenate and integrate the output results of the third multi-head attention unit and the fourth multi-head attention unit to obtain the third concatenated feature, and perform signal detection, recognition, and direction finding of the radiation source based on the third concatenated feature.

[0092] According to the above description, an electronic reconnaissance method provided by the present application adopts multi-channel parallelism to extract features from the collected signals, efficiently extracts the global features and local features of the radiation source signals, and adopts a parallel processing architecture of multiple signal processing channels 131. Each channel corresponds to an independent processing flow of one path of signals, realizing the efficient parallel processing of multiple paths of signals. Combining methods such as the multi-head attention mechanism, shared weight design, residual connection, and layer normalization, it realizes the accurate analysis and complete fusion of signal features, thereby realizing the efficient and accurate processing of radiation source signals in the case of multi-source signal interference, large signal intensity changes, and complex signal frequencies, and quickly and accurately completing the tasks of signal detection, recognition, and direction finding of the radiation source.

[0093] As an optional implementation manner, as Figure 8 shown, in step S12, when extracting features from each path of the collected signals, the steps to obtain the global features of each path of signals are as follows:

[0094] Step S21: Perform Hilbert transform on each path of the collected signals to obtain the Hilbert transform signal;

[0095] Step S22: Concatenate the Hilbert transform signal with the original collected signal to obtain the integrated signal.

[0096] Step S23: Use the long short-term memory network to extract features from the integrated signal to obtain the global features.

[0097] Global features reflect the overall behavior and changing trends of signals, such as the frequency changing trend of signals, the overall fluctuation of power, etc. The obtained global features are used as the data input for subsequent signal processing, providing rich data support for subsequent signal processing.

[0098] As an alternative implementation, as Figure 9 shown, in step S12, feature extraction is performed on each collected signal to obtain the local features of each signal, including the following steps:

[0099] Step S31, splitting the integrated signal into multiple signal segments.

[0100] Step S32, using a long short-term memory network to perform feature extraction on each signal segment.

[0101] Step S33, performing signal splicing on the features extracted by the long short-term memory network to obtain local features.

[0102] Local features reflect the specific behavior and changing trends of signals within different time windows, such as sudden changes in signals within a certain short period of time, short-term fluctuations in specific frequency resolutions, etc. The obtained local features are used as the data input for subsequent signal processing, providing rich data support for subsequent signal processing.

[0103] As an alternative implementation, four signal processing channels 131 for parallel signal processing are set up to calculate the phase difference between the third spliced features of different signal processing channels 131 to obtain the radiation source direction finding result. Further, the third spliced features of different signal processing channels 131 are respectively compared with a preset signal presence judgment criterion to judge whether there is a signal and output the judgment result, realizing signal detection of the radiation source. It also includes respectively matching and classifying the third spliced features of different signal processing channels 131 with the features of different signal types that have been learned, and outputting the modulation type of the signal.

[0104] Four signal processing channels 131 for parallel signal processing are set up, and each passes through an independent processing process corresponding to one path of signal, realizing the simultaneous analysis of the features of multiple signals, ensuring the efficient processing of multiple signals; and through the design of shared weights, the number of model parameters is reduced, the computational complexity is reduced, and the processing efficiency of the system is further improved, so as to realize the rapid processing of the collected signals in the case of multi-source signal interference, large signal strength changes, and complex signal frequencies, and be able to analyze and process the feature information in a timely manner after obtaining the signal features, quickly complete the signal detection, identification, and direction finding tasks of the radiation source, and thus accurately judge the position, type, and threat level of the radiation source.

[0105] It will be understood that the term "exemplary" as used herein means "serving as an example, instance, or illustration". Any embodiment described as "exemplary" is not necessarily preferred or superior to other embodiments and / or does not exclude incorporating features of other embodiments. It should be understood that certain features of the present application described in the context of separate embodiments may also be provided in combination in a single embodiment. Conversely, the various features of the present application described in the context of a single embodiment may also be provided separately or in any suitable combination or as any other described embodiment of the present application.

[0106] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B may mean A or B. The "and / or" herein is merely a relationship describing the associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, "at least one" means one or more, and "a plurality" means two or more. The terms such as "first" and "second" do not limit the quantity and execution order, and the terms such as "first" and "second" do not necessarily mean different.

[0107] The above-disclosed are only the preferred embodiments of the present application, but they are not intended to limit the scope of the rights of the present application. Those of ordinary skill in the art can understand that within the spirit and scope of the present application and the appended claims, changes, modifications, substitutions, combinations, and simplifications should all be equivalent replacement methods and still fall within the scope covered by the invention.

Claims

1. An electronic reconnaissance system, characterized in that, The electronic reconnaissance system includes: An RF module configured to collect signals emitted by radiation sources in multiple channels; A feature extraction module configured to extract features from each collected signal to obtain the global features and local features of each signal respectively; A signal processing module including signal processing channels corresponding to the number of channels of the collected signals. Any one of the signal processing channels includes: A first cross-attention layer including two multi-head attention units with shared weights. Among them, the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features. The query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features; A feature splicing layer including two feature splicing units, and the feature splicing units are configured to splice and integrate the outputs of the first multi-head attention unit and the second multi-head attention unit respectively to obtain a first spliced feature and a second spliced feature; A first normalization layer including a first normalization unit and a second normalization unit, and is configured to perform layer normalization processing on the outputs of the two feature splicing units respectively; A second normalization layer including a third normalization unit and a fourth normalization unit. The input end of the third normalization unit is connected to the output end of the first normalization unit through a fully connected layer and is also connected to the output end of the first normalization unit in a residual connection manner; the input end of the fourth normalization unit is connected to the output end of the second normalization unit through a fully connected layer and is also connected to the output end of the second normalization unit in a residual connection manner; A second cross-attention layer including a third multi-head attention unit and a fourth multi-head attention unit with shared weights. The query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the outputs of the third normalization unit. The query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the outputs of the fourth normalization unit; A second feature splicing layer splices and integrates the output of the second cross-attention layer to obtain a third spliced feature, The signal processing module performs signal detection, identification, and direction finding of the radiation source based on the third spliced feature.

2. The electronic reconnaissance system according to claim 1, wherein The system includes four of the signal processing channels, and the signal processing module further includes a phase difference calculation unit configured to calculate the phase difference between the third spliced features of different signal processing channels to obtain a radiation source direction finding result.

3. The electronic reconnaissance system according to claim 1, wherein The system includes four of the signal processing channels, and the signal processing module further includes a first fully connected layer configured to compare the third spliced features of the four signal processing channels with a preset signal presence judgment criterion respectively and output a judgment result on whether a signal exists.

4. The electronic reconnaissance system according to claim 1, wherein The system includes four of the signal processing channels. The signal processing module further includes a second fully connected layer, which is configured to match and classify the third concatenated features of the four signal processing channels with the features of different signal types that have been learned, and output the modulation type of the signal.

5. The electronic reconnaissance system according to claim 1, wherein the feature extraction module includes: a Hilbert transform unit configured to perform a Hilbert transform on each collected signal to obtain a Hilbert transform signal; a first integration unit configured to concatenate the Hilbert transform signal with the original collected signal to obtain an integrated signal; a first long short-term memory network configured to extract features from the integrated signal to obtain the global features.

6. The electronic reconnaissance system according to claim 5, wherein the feature extraction module further includes: a sliding window cutting unit configured to divide the integrated signal into multiple signal segments; a second long short-term memory network configured to extract features from each of the signal segments; a second integration unit configured to perform signal concatenation on the features extracted by the second long short-term memory network to obtain the local features.

7. An electronic reconnaissance method, characterized in that, The method includes: collecting signals emitted by a radiation source in multiple channels; extracting features from each collected signal to obtain the global features and local features of each signal respectively; processing each signal in a multi-channel parallel manner, and any signal processing channel processes the global features and local features through two weight-sharing multi-head attention units, wherein the query input of the first multi-head attention unit is the local feature, and the key input and value input are the global features, and the query input of the second multi-head attention unit is the global feature, and the key input and value input are the local features; concatenating and integrating the outputs of the first multi-head attention unit and the second multi-head attention unit respectively through a feature concatenation unit to obtain a first concatenated feature and a second concatenated feature; performing layer normalization processing on the outputs of the two feature concatenation units respectively through a first normalization unit and a second normalization unit; transmitting the output of the first normalization unit to a third normalization unit and the output of the second normalization unit to a fourth normalization unit respectively through a fully connected layer and a residual connection, and the third normalization unit and the fourth normalization unit perform layer normalization processing on the received signals respectively; performing weighted fusion on the outputs of the third normalization unit and the fourth normalization unit through a weight-sharing third multi-head attention unit and a fourth multi-head attention unit, wherein the query input of the third multi-head attention unit is the output of the fourth normalization unit, and the key input and value input are the output of the third normalization unit, and the query input of the fourth multi-head attention unit is the output of the third normalization unit, and the key input and value input are the output of the fourth normalization unit; The output results of the third multi-head attention unit and the fourth multi-head attention unit are spliced and integrated to obtain a third spliced feature, and signal detection, recognition, and direction finding of the radiation source are performed based on the third spliced feature.

8. The electronic reconnaissance method according to claim 7, wherein Four signal processing channels for parallel processing of signals are set, and the phase difference between the third spliced features of different signal processing channels is calculated to obtain the radiation source direction finding result.

9. The electronic reconnaissance method according to claim 7, wherein Perform Hilbert transform on each path of the collected signals to obtain Hilbert transform signals; Splice the Hilbert transform signals and the original signals collected to obtain integrated signals; Use a long short-term memory network to extract features from the integrated signals to obtain the global features.

10. The electronic reconnaissance method according to claim 9, wherein The integrated signals are segmented into multiple signal segments; Use a long short-term memory network to extract features from each of the signal segments; The features extracted by the long short-term memory network are signal-spliced to obtain the local features.

Images