An RCS System Simulation and Testing Method and System Based on a Deep Neural Network

Abstract

The present invention discloses an RCS system simulation test method and system based on a deep neural network, including segmenting the collected RCS target echo feature data to obtain a time series feature sequence; constructing a dual-branch deep neural network structure to process the time series feature sequence, fusing the features extracted by the main and auxiliary branches to obtain an RCS feature representation; setting a cross-scale feature correction unit in the dual-branch deep neural network structure to calculate the cross-correlation coefficient between the main and auxiliary branch feature maps and dynamically correct the main and auxiliary branch features; generating high-frequency and low-frequency RCS simulation features based on the output of the correction unit and performing adaptive fusion to obtain RCS system simulation features; calculating the simulation error between the RCS system simulation features and the reference features, and optimizing the network parameters. The present invention significantly improves the adaptive ability and stability of the measurement system, and is of great significance for improving the overall level of performance evaluation of the electronic countermeasure system.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing and test evaluation, and particularly to an RCS system simulation test method and system based on a deep neural network. Background Art

[0002] As an important test means in modern electronic countermeasure systems, the simulation accuracy and reliability of radar cross-section (RCS) measurement directly affect the evaluation results of weapon equipment performance. With the continuous development of stealth technology for modern combat platforms, the accuracy requirements for RCS measurement systems are getting higher and higher. Existing RCS simulation test systems mainly focus on the acquisition and processing efficiency of measurement data, but there are still obvious deficiencies in the quality evaluation and dynamic optimization of measurement results.

[0003] Currently, RCS measurement systems usually evaluate measurement quality using a single index, such as signal-to-noise ratio or measurement error, and it is difficult to comprehensively reflect the scattering characteristics of complex targets at different postures and frequencies. Especially when dealing with multi-targets or complex structure targets, there is a lack of a detailed evaluation mechanism for local scattering characteristics. At the same time, existing systems mostly adopt static measurement strategies and lack the ability of real-time evaluation and parameter adjustment during the measurement process, which is likely to cause problems such as unstable measurement accuracy or loss of local features.

[0004] More importantly, existing technologies rarely consider the mutual correlation between the scattering characteristics of different parts of the target and lack an effective monitoring mechanism for the change of data correlation during the measurement process. In the RCS measurement of complex targets, the issue of maintaining the organizational relationship between local scattering characteristics is often ignored, which may lead to the measurement results not being able to truly reflect the scattering characteristics of the target. In addition, existing systems mostly adopt fixed measurement parameters and are difficult to adaptively adjust measurement strategies according to different target characteristics and measurement environments, which severely restricts the improvement of measurement accuracy. These technical deficiencies not only affect the accuracy and reliability of RCS measurement but also limit the accuracy of the performance evaluation of electronic countermeasure systems. Summary of the Invention

[0005] In view of the problems existing in the existing RCS system simulation test method based on a deep neural network, the present invention proposes an RCS system simulation test method and system based on a deep neural network.

[0006] Therefore, the problem to be solved by the present invention is to improve the reliability test of the measurement results of the RCS simulation test system.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] First aspect, an embodiment of the present invention provides a simulation test method for an RCS system based on a deep neural network, which includes: collecting RCS target echo feature data, segmenting the RCS target echo feature data according to a preset time window length to obtain a time series feature sequence; constructing a dual-branch deep neural network structure based on the time series feature sequence, the main branch uses a standard convolutional layer to process the time series feature sequence, the auxiliary branch is provided with a spectrum reconstruction module, and the features extracted by the main branch and the auxiliary branch are fused to obtain an RCS feature representation; setting a cross-scale feature correction unit in the dual-branch deep neural network structure, and the cross-scale feature correction unit dynamically corrects the main and auxiliary branch features by calculating the cross-correlation coefficient between the feature maps of the main branch and the auxiliary branch; based on the output of the cross-scale feature correction unit, generating high-frequency RCS simulation features and low-frequency RCS simulation features through the main branch decoder and the auxiliary branch decoder respectively, and performing adaptive fusion using the cross-correlation coefficient matrix to obtain RCS system simulation features; using the RCS feature representation as a reference feature, calculating the simulation error between the RCS system simulation features and the reference feature, and optimizing the network parameters based on the simulation error to improve the simulation accuracy of the RCS system.

[0009] As a preferred solution of the simulation test method for the RCS system based on the deep neural network of the present invention, wherein: collecting the RCS target echo feature data, performing band-pass filtering on the original target echo data to obtain a filtered echo signal, and performing complex resampling to obtain a discrete sampling sequence; applying a window function to the discrete sampling sequence for segmental windowing, and performing a fast Fourier transform on the windowed signal sequence to obtain spectral features, and extracting amplitude spectrum and phase spectrum information from the spectral features to form a time series feature sequence.

[0010] As a preferred solution of the simulation test method for the RCS system based on the deep neural network of the present invention, wherein: the main branch of the dual-branch deep neural network structure is used to process the time series feature sequence, and the auxiliary branch of the dual-branch deep neural network structure is used to process frequency domain components, and the frequency domain components are the amplitude spectrum and phase spectrum information; the main branch uses a standard convolutional layer to process the time series feature sequence; and a residual connection is added every two convolutional layers to form a residual block structure; the residual block structure is a dual-path structure, including a main path and a shortcut connection path; adding the features processed by the main path and the features of the shortcut connection path element-wise, and the combined features are then passed through the ReLU activation function to obtain the final output of the residual block. Collecting the feature maps of the outputs of each residual block, the feature maps include shallow feature maps and deep feature maps; using an adaptive weight mechanism to perform feature fusion on the feature maps to obtain a fused feature representation; performing dimensionality compression processing on the fused feature representation to obtain a main branch feature vector with a fixed dimension 。

[0011] As a preferred solution of the RCS system simulation test method based on a deep neural network according to the present invention, wherein: the sub-branch receives amplitude spectrum and phase spectrum information, inputs the amplitude spectrum into a frequency-domain feature extraction module composed of L one-dimensional convolutional layers to generate a frequency distribution feature; introduces a compensation mechanism for the phase spectrum, and based on the compensated phase spectrum and the original amplitude spectrum reconstructs the frequency-domain signal :

[0012]

[0013] wherein, is the imaginary unit, is the compensated phase spectrum; converts the reconstructed frequency-domain signal into a feature vector through a feature mapping network composed of two fully connected layers ; concatenates the main-branch feature vector and the branch feature vector on the feature dimension to form a fused feature vector; performs dimensionality reduction mapping on the fused feature vector to obtain an RCS feature representation .

[0014] As a preferred solution of the RCS system simulation test method based on a deep neural network according to the present invention, wherein: the cross-scale feature correction unit includes a cross-correlation calculation module, a weight generation module, and a feature correction module. The cross-correlation calculation module is used to calculate the correlation matrix between the features of the two branches, the weight generation module is used to generate the weight factor of the feature channels, and the feature correction module is used to optimize and correct the features; the calculation of the correlation matrix is performed by extracting the main and auxiliary branch feature maps, and adjusting them to the same feature space through projection transformation to calculate the cross-correlation coefficient matrix , specifically including: the main-branch feature map is extracted from each residual block , wherein , are the height and width of the feature map respectively, is the number of feature channels, represents the th residual block level; the auxiliary-branch feature map is obtained from the middle layer of the feature mapping network ; rearranges the main-branch feature map and converts it into a matrix form ; performs dimensionality adjustment and normalization processing on the auxiliary-branch features and converts them into a matrix form compatible with the main-branch features ; calculates the cross-correlation coefficient matrix between the main and auxiliary branch features :

[0015]

[0016] Among them, is the cross-correlation coefficient matrix, indicating the direct correlation between features, is the autocorrelation matrix of the main branch features, is the autocorrelation matrix of the auxiliary branch features; the cross-correlation coefficient matrix is normalized to obtain the main branch feature weight factor ; the weight factor is used to characterize the contribution degree of each feature channel of the main branch to the RCS feature representation; the auxiliary branch feature weight factor is obtained through column normalization; the weight factor is used to characterize the contribution degree of each feature channel of the auxiliary branch to the RCS feature representation; the feature correction module applies the weight factor to the main branch features to obtain the corrected feature map ; the weight factor is applied to the auxiliary branch features to obtain the corrected feature map .

[0017] As a preferred solution of the RCS system simulation test method based on a deep neural network according to the present invention, wherein: based on the corrected feature map it is reconstructed into a high-frequency RCS simulation feature through a decoder; based on the corrected feature map , it is reconstructed into a low-frequency RCS simulation feature through an auxiliary branch decoder; the high-frequency RCS simulation feature and the low-frequency RCS simulation feature are subjected to dimension unification processing and normalization to generate a high-frequency feature weight matrix and a low-frequency feature weight matrix , the weight matrices are respectively multiplied element by element with the corresponding feature maps, and element-by-element addition operations are performed on the weighted high-frequency and low-frequency features to obtain the RCS system simulation feature .

[0018] As a preferred solution of the RCS system simulation test method based on a deep neural network according to the present invention, wherein: based on the RCS system simulation feature and the RCS feature representation, the feature reconstruction error between the two is calculated, specifically: calculating the feature reconstruction error between the RCS system simulation feature and the RCS feature representation; for the high-frequency RCS feature, calculating the matching degree with the high-frequency components in the RCS feature representation ; for the low-frequency RCS feature, evaluating the consistency with the low-frequency components in the RCS feature representation ; analyzing the cross-correlation coefficient matrix Evaluate the fusion quality based on the changes, and obtain the standardized branch feature error by comprehensively considering the matching degree of high-frequency RCS features, the consistency of low-frequency RCS features, and the fusion quality. Calculate the total error based on the feature reconstruction error weight and the branch feature error weight, expressed as:

[0019]

[0020] Where is the feature reconstruction error weight, is the branch feature error weight.

[0021] As a preferred solution of the RCS system simulation test method based on a deep neural network according to the present invention, wherein: based on the change trends of various errors, optimize the parameter settings, including: if entering the main branch optimization, first optimize the parameters of the residual block convolutional layer, adjust the parameters of the adaptive weight mechanism, update the parameters of the main branch decoder, and complete the enhancement of the main feature extraction ability; if the main branch optimization is completed, optimize the parameters of the feature mapping network in the auxiliary branch, update the parameters of the auxiliary branch decoder, and improve the auxiliary feature extraction ability; if the total error is lower than the preset threshold, or the error change in consecutive rounds is less than the convergence threshold, or the maximum number of iterations is reached, or the validation set error has not improved for consecutive rounds, then terminate the optimization process and complete the network parameter optimization.

[0022] In a second aspect, an embodiment of the present invention provides an RCS system simulation test system based on a deep neural network, which includes: a data acquisition module for collecting RCS target echo feature data, segmenting the RCS target echo feature data according to a preset time window length to obtain a time series feature sequence; a feature representation module for constructing a double-branch deep neural network structure based on the time series feature sequence, the main branch using a standard convolutional layer to process the time series feature sequence, the auxiliary branch setting a spectrum reconstruction module to fuse the features extracted by the main branch and the auxiliary branch to obtain an RCS feature representation; a dynamic correction module for setting a cross-scale feature correction unit in the double-branch deep neural network structure, and dynamically correcting the main and auxiliary branch features by calculating the cross-correlation coefficient between the feature maps of the main branch and the auxiliary branch; a simulation optimization module for respectively generating high-frequency RCS simulation features and low-frequency RCS simulation features, and performing adaptive fusion to obtain RCS system simulation features, using the RCS feature representation as a reference feature, calculating the simulation error between the RCS system simulation features and the reference feature, and optimizing the network parameters based on the simulation error to improve the RCS system simulation accuracy.

[0023] The beneficial effects of the present invention are as follows: by applying deep neural network technology to the RCS simulation test system, an intelligent measurement quality evaluation framework is constructed. The convolutional neural network and recurrent neural network are used to extract and analyze the features of measurement data, and a multi-dimensional evaluation mechanism including mean proximity, fluctuation consistency, and structural correlation is established, realizing a comprehensive evaluation of the quality of measurement data. Based on the feature extraction ability of the deep learning model, the system can accurately identify abnormal patterns in the measurement results and adaptively adjust the measurement strategy through reinforcement learning methods. At the same time, the continuous learning and optimization ability of the deep learning model enables the system to continuously improve the evaluation accuracy and robustness, providing a new technical path for the intelligent upgrade of the RCS test system, significantly enhancing the adaptive ability and stability of the measurement system, and being of great significance for improving the overall level of performance evaluation of electronic countermeasure systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0025] Figure 1 It is a flowchart of the RCS system simulation test method based on a deep neural network. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0028] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0029] Embodiment 1

[0030] Referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a simulation test method for an RCS system based on a deep neural network, including:

[0031] S1: Collect RCS target echo feature data, segment the RCS target echo feature data according to a preset time window length, perform a fast Fourier transform on the windowed signal sequence to obtain spectral features, and extract amplitude spectrum and phase spectrum information from the spectral features to form a time series feature sequence.

[0032] The collection of RCS target echo feature data is processed by collecting data using a multi-angle radar, including:

[0033] Using a phased array radar system with a working frequency of to scan the RCS target within the measurement angle range at an angle interval of , and obtain the original target echo data, where the angle interval is less than 1 / 30 of the measurement angle range;

[0034] Perform band-pass filtering on the original target echo data, and the bandwidth of the band-pass filtering satisfies 0.01 ≤ ≤0.1 , and obtain the filtered echo signal;

[0035] Perform complex resampling on the filtered echo signal, and the sampling rate of the complex resampling is not less than 2 times the bandwidth , and obtain a discrete sampling sequence.

[0036] Perform segmented windowing processing on the discrete sampling sequence by applying a window function, including:

[0037] Select a window function according to the frequency characteristics of the discrete sampling sequence, specifically by adjusting the ratio of the main lobe width to the side lobe amplitude of the window function to achieve a balance between frequency resolution and dynamic range;

[0038] The window function can smoothly transition to zero at the time domain endpoints, suppress the spectral leakage effect caused by truncation, and improve the accuracy of spectral analysis;

[0039] Perform segmented processing on the windowed sequence using the overlap-add method, and maintain the continuity of the signal by adjusting the length of the overlapping region. The length of the overlapping region is determined according to the time-varying characteristics of the signal, and the adjacent window overlap rate is between 30% - 70%, and obtain a windowed signal sequence.

[0040] Perform a fast Fourier transform on the windowed signal sequence to obtain spectral features, specifically:

[0041] Zero-pad the windowed signal sequence to \(N\) sampling points, where \(N\) is an integer power of 2 and \(N\) is greater than the number of sampling points of the windowed signal sequence;

[0042] Perform butterfly operation decomposition on the zero-padded signal sequence, and decompose the \(N\)-point Fourier transform into \(L\) levels of 2-point Fourier transforms;

[0043] Calculate the rotation factors for each level of 2-point Fourier transform. The rotation factor is , where is the frequency index;

[0044] Perform butterfly operations according to the decimation-in-time method to obtain spectral features, and the spectral features include non-redundant frequency points.

[0045] Extract amplitude spectrum and phase spectrum information from the spectral features to form a time-series feature sequence, and the feature dimension of the time-series feature sequence does not exceed half of the number of points of the fast Fourier transform.

[0046] Among them, the amplitude spectrum reflects the energy distribution intensity of the signal at different frequencies, and the phase spectrum reflects the relative phase relationship of each frequency component; in order to construct the time-series feature sequence, it is necessary to combine the amplitude spectrum and phase spectrum information. The combination process is to combine the amplitude value and the corresponding phase angle at each frequency point to form a complete frequency-domain representation, and through the inverse Fourier transform, this complete frequency-domain representation is converted back to the time domain. The conversion process can be expressed as: all frequency components are superimposed according to their respective amplitudes and phases, and each frequency component will contribute to the final time-domain signal. Larger-amplitude frequency components will have a more significant impact on the time-domain signal, and the phase determines the precise position of these components in time.

[0047] After the conversion is completed, a new time-series feature sequence is obtained. This sequence maintains the same time length as the original windowed signal, but it contains optimized information after frequency-domain analysis and reconstruction.

[0048] S2: Construct a dual-branch deep neural network structure based on the time-series feature sequence and frequency-domain components. Among them, the main branch uses standard convolutional layers to process the time-series feature sequence, the auxiliary branch is provided with a spectrum reconstruction module, the spectrum reconstruction module reconstructs the frequency-domain information by introducing a phase compensation mechanism, and fuses the features extracted by the main branch and the auxiliary branch to obtain the RCS feature representation.

[0049] Through the time-frequency domain decomposition of the signal, a time-series feature sequence and frequency-domain components are obtained, and the frequency-domain components are the amplitude spectrum and the phase spectrum.

[0050] Based on the decomposed time-series feature sequence and frequency-domain components, construct a dual-branch deep neural network structure to process time-domain and frequency-domain information respectively, specifically including:

[0051] Construct the main branch of the dual-branch deep neural network structure, including:

[0052] Extract features from the time series feature sequence through N layers of standard one-dimensional convolutional layers. The convolutional kernel size of each convolutional layer is , the number of convolutional kernels is , the stride is , and the padding is ;

[0053] Adopt an increasing convolutional kernel size from the 1st layer to the N / 2th layer to capture local features of different scales;

[0054] Adopt a decreasing convolutional kernel size from the (N / 2 + 1)th layer to the Nth layer to achieve layer-by-layer refinement of features;

[0055] After the output of each convolutional layer, concatenate a batch normalization layer to eliminate data distribution shift. Specifically, standardize the feature map output by the convolutional layer. First, calculate the mean and variance of each batch of data, and perform normalization by subtracting the mean and dividing by the variance, and introduce learnable scaling factor and offset factor to increase the expressive ability of the model.

[0056] Add residual connections every two convolutional layers to form a residual block structure, alleviating the gradient vanishing problem in deep networks. The residual block adopts a dual-path structure including a main path and a shortcut connection path;

[0057] On the main path, the input features first pass through the first convolutional layer for feature extraction, then sequentially pass through the batch normalization layer to eliminate data distribution shift, and then pass through the ReLU activation function to introduce non-linear features. The processed features continue to pass through the second convolutional layer for further extraction and then pass through the batch normalization layer again. At the same time, the input features are directly passed to the end of the residual block through the shortcut connection path. When the dimensions of the input features do not match those of the output features of the main path, a 1×1 convolutional layer is needed to adjust the dimensions of the features in the shortcut connection. Add the features processed on the main path and the features in the shortcut connection path element-wise, and the combined features pass through the ReLU activation function to obtain the final output of the residual block.

[0058] The main path can learn new feature transformations, and the shortcut connection can retain the original feature information, thus better transmitting gradient information in deep networks and effectively alleviating the gradient vanishing problem. The double-convolution structure on the main path enhances the feature extraction ability, and the shortcut connection ensures the effective transmission of information. The combination of the two improves the learning ability and training stability of the entire network.

[0059] Collect the feature maps output by each residual block. The feature maps contain feature information at different levels. Shallow feature maps mainly contain local detail information, while deep feature maps contain more high-level semantic information. To make full use of multi-level features, an adaptive weight mechanism is adopted for feature fusion: calculate the importance weights for the feature maps at each level, and the importance weights are automatically adjusted by learnable parameters; weight the features at different levels according to the calculated importance weights, so that more important features obtain higher weight values; perform element-wise addition on the weighted feature maps to obtain the fused feature representation.

[0060] To obtain a feature vector with a fixed dimension, dimension compression processing is performed on the fused feature representation. Specifically, the temporal dimension is compressed through a global pooling operation, and the entire temporal feature is compressed into a single feature value. The global pooling process can perform both max-pooling and average-pooling operations simultaneously, and adaptively fuse the results of the two pooling operations through learnable parameters to retain the most representative feature information. Finally, a main-branch feature vector with a fixed dimension is obtained. , It not only retains the multi-level feature information but also has a fixed dimension.

[0061] Construct the auxiliary branch of the dual-branch deep neural network structure, including:

[0062] Receive the output amplitude spectrum and phase spectrum, input the amplitude spectrum into the frequency-domain feature extraction module composed of L one-dimensional convolutional layers to generate frequency distribution features;

[0063] Introduce a compensation mechanism for the phase spectrum, and the compensated phase spectrum satisfies:

[0064]

[0065] where is the original phase spectrum, is the compensation amplitude coefficient, is the compensation frequency coefficient, is the compensation phase offset, , , are all learnable parameters.

[0066] Based on the compensated phase spectrum and the original amplitude spectrum reconstruct the frequency-domain signal :

[0067]

[0068] where, is the imaginary unit, and the reconstructed signal enhances the expression of phase features.

[0069] Convert the reconstructed frequency-domain signal into a feature vector through a feature mapping network composed of two fully-connected layers , the output dimension of the feature mapping network is the same as that of the main branch feature vector , specifically including:

[0070] Pass the reconstructed frequency-domain signal through a feature mapping network composed of two fully-connected layers. First, the reconstructed frequency-domain signal enters the first fully-connected layer as input. The number of neurons in this layer is greater than the input dimension, playing a role in feature expansion. After the first fully-connected layer, a BatchNorm normalization layer and a ReLU activation function are connected, introducing non-linear transformation ability while maintaining the stability of the data distribution.

[0071] Then, the output features of the first layer continue to be fed into the second fully-connected layer. The number of neurons in this layer is precisely set to be the same dimension as the main branch feature vector . The second fully-connected layer is also equipped with a BatchNorm normalization layer, but finally uses the Tanh activation function instead of ReLU because the output range of the Tanh function is between [-1,1], which helps with the normalized representation of features, making the generated feature vector have a similar numerical distribution range to the main branch feature vector .

[0072] Concatenate the main branch feature vector and the branch feature vector on the feature dimension to form a fused feature vector.

[0073] Perform a dimensionality reduction mapping on the fused feature vector through a fully-connected layer to obtain the final RCS feature representation , expressed as:

[0074]

[0075] where is the weight matrix, is the bias vector, represents the feature concatenation operation.

[0076] S3: Set a cross-scale feature correction unit in the dual-branch deep neural network structure. The cross-scale feature correction unit dynamically corrects the main and auxiliary branch features by calculating the cross-correlation coefficient between the main branch and auxiliary branch feature maps.

[0077] The cross-scale feature correction unit includes a cross-correlation calculation module, a weight generation module, and a feature correction module. The cross-correlation calculation module is used to calculate the correlation matrix between the two branch features, the weight generation module is used to generate the weight factors for the feature channels, and the feature correction module is used to perform weighted optimization on the features.

[0078] The cross - correlation calculation module calculates the correlation matrix between the features of two branches, including:

[0079] Extract the feature maps of the main and auxiliary branches:

[0080] Main branch: Extract the feature map from each residual block , where and are the height and width of the feature map respectively, is the number of feature channels, represents the th residual block level. The features of each residual block retain the time - domain dynamic information of different scales;

[0081] Auxiliary branch: Obtain the feature map from the middle layer of the feature mapping network , which contains the frequency - domain information after phase compensation.

[0082] Since the features of the main branch come from the residual blocks and the features of the auxiliary branch come from the feature mapping network, and their feature representation forms are different, feature alignment is required; Align the main - branch feature map and the auxiliary - branch feature map to the same feature space through projection transformation to ensure that the feature dimensions match. Specifically:

[0083] Rearrange the main - branch feature map and convert it into matrix form ;

[0084] Adjust the dimensions and normalize the features of the auxiliary branch, and convert them into a matrix form compatible with the main - branch features ;

[0085] Ensure that the features of the two branches are represented in the same feature space through projection transformation.

[0086] Calculate the cross - correlation coefficient matrix between the features of the main and auxiliary branches :

[0087]

[0088] Among them, is the cross - correlation coefficient matrix, represents the direct correlation between features, represents the autocorrelation matrix of the main - branch features, represents the autocorrelation matrix of the auxiliary - branch features.

[0089] The element in the cross - correlation coefficient matrix represents the rd main - branch feature channel and the th auxiliary - branch feature channel's correlation strength.

[0090] Each row of the matrix represents the correlation distribution of a main branch channel with all secondary branch channels;

[0091] Each column of the matrix represents the correlation distribution of a secondary branch channel with all main branch channels;

[0092] By normalizing the cross - correlation coefficient matrix the main - branch feature weight factor ;

[0093] The weight factor is used to characterize the contribution degree of each feature channel of the main branch to the RCS feature representation;

[0094] The secondary - branch feature weight factor is obtained through column normalization;

[0095] The weight factor is used to characterize the contribution degree of each feature channel of the secondary branch to the RCS feature representation.

[0096] Applying the weight factor to the main - branch features, the corrected feature map is obtained. The correction process is to process the weight factor through the sigmoid function, map it to the interval [0, 1], and then multiply it with the feature map for correction. The correction process ensures highlighting the key information in the time - domain features and suppressing the secondary features.

[0097] Applying the weight factor to the secondary - branch features, the corrected feature map is obtained. After the correction process also uses the sigmoid function to process the weight factor and then multiplies it with the feature map for correction. The correction process ensures the coordination between the frequency - domain features and the time - domain features.

[0098] S4. Based on the output of the cross - scale feature correction unit, the high - frequency RCS simulation features and the low - frequency RCS simulation features are respectively generated through the main - branch decoder and the secondary - branch decoder, and an adaptive fusion is performed using the cross - correlation coefficient matrix to obtain the complete RCS system simulation features.

[0099] The corrected feature map contains the key time - domain information of the target, but due to the previous down - sampling and convolution operations, the spatial resolution is low. It is necessary to reconstruct it into high - frequency RCS simulation features through the decoder.

[0100] The decoding process adopts a multi-layer transposed convolution structure, and gradually improves the spatial resolution of the feature map through transposed convolution operations.

[0101] Specifically, First, it is input into the first transposed convolution layer, and the features are mapped to a higher-resolution space through transposed convolution operations. Transposed convolution expands the size of the feature map by inserting zero values between the feature maps and performing convolution operations. At the same time, each layer of transposed convolution is followed by a non-linear activation function to enhance the expression ability of the features. To avoid information loss during the feature transmission process, residual connections are added between adjacent transposed convolution layers to retain the original feature information.

[0102] Since each layer of transposed convolution can learn feature patterns at different scales, the shallower layers mainly reconstruct the basic structure, and the deeper layers can recover more fine-grained local details. As the features are gradually transmitted through the multi-layer transposed convolution structure, the spatial resolution is continuously improved, and the detailed information of the features is gradually restored. The finally output high-frequency RCS simulation features not only retain the key time-domain information in, but also contain rich high-frequency details.

[0103] Similarly, after obtaining the corrected auxiliary branch features they are reconstructed into low-frequency RCS simulation features through the auxiliary branch decoder. Since low-frequency features reflect the overall structural characteristics of the target, the decoding process needs to pay attention to the global consistency of the features.

[0104] The auxiliary branch decoder also adopts a multi-layer transposed convolution structure, but it is different from the main branch decoder. First, After the first transposed convolution, a convolution kernel with a larger size is used for feature mapping, so that each output position can obtain a larger range of input information, which is beneficial to maintaining the integrity of the features. At the same time, global average pooling operations are introduced between the transposed convolution layers to extract and fuse global feature information and enhance the overall expression ability of the features.

[0105] During the feature transmission process, the spatial resolution of the feature map is improved through layer-by-layer transposed convolution operations. A normalization layer is added after each layer of transposed convolution to stabilize the feature distribution and ensure the stability of the reconstruction process. In addition, in order to better maintain the continuity of the low-frequency features, skip connections are used between adjacent layers to fuse the basic features of the lower layers and the semantic features of the higher layers, and output low-frequency RCS simulation features that retain the main structural information and overall contour features of the target.

[0106] The high-frequency RCS simulation features and the low-frequency RCS simulation features are processed to unify their dimensions, so that the two have the same spatial resolution and the same number of feature channels; the Softmax function is used for the cross-correlation coefficient matrix Normalize the elements in it, map the correlation coefficients to the interval [0, 1], and generate a high-frequency feature weight matrix and a low-frequency feature weight matrix , and ensure that the weight matrix satisfies the constraint conditions;

[0107] Perform an element-wise multiplication operation between the weight matrix and the corresponding feature map respectively, and perform an element-wise addition operation on the weighted high-frequency and low-frequency features to obtain the complete RCS system simulation feature , which is expressed as:

[0108]

[0109] where represents the high-frequency feature, represents the low-frequency feature, and ⊙ represents the element-wise multiplication operation.

[0110] S5. Use the RCS feature representation as the reference feature, calculate the simulation error between the complete RCS system simulation feature and the reference feature, and optimize the network parameters based on the simulation error to improve the RCS system simulation accuracy.

[0111] Based on the RCS system simulation feature and the RCS feature representation calculate the feature reconstruction error between the two, specifically:

[0112] First, calculate the Euclidean distance error between the simulation feature and the RCS feature representation , evaluate the overall reconstruction effect, including:

[0113] Set the mean Euclidean distance , between different category samples in the labeled dataset as the preset convergence threshold, and < ,

[0114] If < , then the reconstructed feature has achieved a match with the reference feature, and transfer to the local feature optimization stage;

[0115] If ≤ < , then the reconstructed feature only obtains the target main feature, and continue to optimize the feature extraction network;

[0116] If ≥ , then there is an essential difference between the reconstructed feature and the reference feature, and reconstruct the feature extraction scheme.

[0117] Further analyze the direction consistency of the two feature vectors through cosine similarity to ensure the alignment of the feature space, including:

[0118] If the cosine similarity is close to 1, it indicates that the reconstructed feature is highly consistent with the original feature in direction, and it is confirmed that the feature space is completely aligned;

[0119] If the cosine similarity is at a medium level, it indicates that the directions of the feature vectors are basically aligned, and fine-tuning optimization is required;

[0120] If the cosine similarity is close to 0, it indicates that there is a significant deviation in the directions of the feature vectors, and the reconstruction strategy needs to be readjusted.

[0121] Based on the feature structure similarity analysis, evaluate the fidelity of the reconstructed feature in the local area, including:

[0122] If the difference in regional averages is small, it indicates that the local feature intensities match, ensuring the consistency of local feature levels;

[0123] If the regional fluctuation degrees are similar, it indicates that the local change patterns are consistent, and it is confirmed that the local dynamic characteristics are maintained;

[0124] If the regional correlation is strong, it indicates that the local structural relationships are stable, verifying the integrity of the detailed features.

[0125] Normalize the above three errors to obtain the standardized feature reconstruction error .

[0126] For the high-frequency RCS features generated by the main branch, calculate the matching degree with the high-frequency components in the RCS feature representation including:

[0127] Perform frequency domain transformation on the high-frequency RCS features and the RCS feature representation Extract the amplitude distribution of the high-frequency band, obtain the high-frequency phase information, get the high-frequency feature components to be evaluated and the reference high-frequency components, and evaluate the error degrees of the high-frequency feature components and the reference high-frequency components in terms of amplitude difference, phase shift, and spectral structure to obtain the matching degree of the high-frequency RCS.

[0128] For the low-frequency RCS features generated by the auxiliary branch, evaluate the consistency with the low-frequency components in the RCS feature representation including:

[0129] Perform frequency domain transformation on the low-frequency RCS features and the RCS feature representation Perform frequency domain transformation, extract the amplitude distribution of the low-frequency band, obtain the low-frequency phase information, get the low-frequency feature components to be evaluated and the reference low-frequency components, evaluate the consistency deviation of the low-frequency feature components and the reference low-frequency components in terms of baseline offset degree, trend change consistency, and energy distribution characteristics, and evaluate the consistency of the low-frequency RCS characteristics.

[0130] Analyze the cross-correlation coefficient matrix during the feature fusion process, and evaluate the fusion quality, including:

[0131] Analysis of diagonal element stability: By comparing the change in the autocorrelation intensity of the main features at consecutive time steps, when the change amplitude continuously remains less than the preset threshold, it indicates that the self-expression of the features in each dimension tends to be stable;

[0132] Analysis of non-diagonal element convergence: Monitor the degree of interaction correlation between different features. When the change rate of the interaction intensity decreases and remains at a low level, it shows that the information exchange between features has reached equilibrium, and the feature fusion enters a stable stage;

[0133] Analysis of matrix structure consistency: Compare the overall correlation structure of adjacent time steps. When the structural difference of the matrix continuously decreases and remains at a low level, it indicates that the feature fusion network reaches a coordinated and consistent state, and the fusion quality is in the optimal range.

[0134] Based on the matching degree of the high-frequency RCS characteristics, the consistency of the low-frequency RCS characteristics, and the fusion quality, obtain the standardized branch feature error .

[0135] Define the adaptive weight coefficient to dynamically balance the contributions of the feature reconstruction error and the branch feature error. The weight coefficient is automatically adjusted according to the current training stage and the error distribution. Moreover, the weight of the feature reconstruction error gradually increases during training to ensure the final reconstruction quality; the weight of the branch feature error is larger in the initial stage of training to ensure the accurate extraction of the basic features.

[0136] Calculate the total error based on the feature reconstruction error weight and the branch feature error weight, expressed as:

[0137]

[0138] where is the feature reconstruction error weight, is the branch feature error weight.

[0139] Record the change trend of various errors in each round of iteration. According to the error analysis results, automatically adjust the focus of the next round of optimization. When a certain type of error is significantly higher than other errors, correspondingly increase its weight coefficient, and establish an error change curve to guide the dynamic adjustment of the optimization process.

[0140] Optimize the parameter settings based on the changing trends of various errors, including:

[0141] If starting network training, set the initial learning rate and the maximum number of iterations , thereby initializing the Adam optimizer and determining the learning rate decay strategy.

[0142] If entering the main branch optimization, first optimize the parameters of the residual block convolutional layer, adjust the parameters of the adaptive weight mechanism, update the parameters of the main branch decoder, and complete the enhancement of the main feature extraction ability;

[0143] If the main branch optimization is completed, optimize the parameters of the feature mapping network in the auxiliary branch and update the parameters of the auxiliary branch decoder to enhance the auxiliary feature extraction ability;

[0144] If the main and auxiliary branch optimizations are completed, then optimize the feature correction unit, calculate and optimize the cross-correlation coefficient matrix , dynamically adjust the feature weight factor, and achieve precise control of feature fusion.

[0145] If each iteration is completed, calculate the current overall error, update the parameters using the Adam optimizer, and dynamically adjust the learning rate according to the error change to ensure the efficiency of the optimization process;

[0146] If the total error is lower than the preset threshold, or the error change is less than the convergence threshold for consecutive rounds, or the maximum number of iterations is reached , or the validation set error has not improved for consecutive rounds, then terminate the optimization process and complete the network parameter optimization.

[0147] In summary, the present invention applies deep neural network technology to the RCS simulation test system to construct an intelligent measurement quality evaluation framework. The convolutional neural network and recurrent neural network are used to extract and analyze the measurement data, and a multi-dimensional evaluation mechanism including mean proximity, fluctuation consistency, and structural correlation is established to achieve a comprehensive evaluation of the measurement data quality. Based on the feature extraction ability of the deep learning model, the system can accurately identify abnormal patterns in the measurement results and adaptively adjust the measurement strategy through the reinforcement learning method. At the same time, the continuous learning and optimization ability of the deep learning model enables the system to continuously improve the evaluation accuracy and robustness, providing a new technical path for the intelligent upgrade of the RCS test system, significantly enhancing the adaptive ability and stability of the measurement system, and having important significance for improving the overall level of the performance evaluation of the electronic countermeasure system.

[0148] Embodiment 2

[0149] On the basis of the first embodiment, this embodiment further provides an RCS system simulation test system based on a deep neural network, including:

[0150] A data acquisition module, configured to acquire RCS target echo feature data, segment the RCS target echo feature data according to a preset time window length, and obtain a time series feature sequence;

[0151] A feature representation module, which constructs a two-branch deep neural network structure based on the time series feature sequence. The main branch uses a standard convolutional layer to process the time series feature sequence, and the auxiliary branch is provided with a spectrum reconstruction module to fuse the features extracted by the main branch and the auxiliary branch to obtain an RCS feature representation;

[0152] A dynamic correction module, which sets a cross-scale feature correction unit in the two-branch deep neural network structure, and dynamically corrects the main and auxiliary branch features by calculating the cross-correlation coefficient between the feature maps of the main branch and the auxiliary branch;

[0153] A simulation optimization module, configured to generate high-frequency RCS simulation features and low-frequency RCS simulation features through the main branch decoder and the auxiliary branch decoder respectively, perform adaptive fusion to obtain RCS system simulation features, use the RCS feature representation as a reference feature, calculate the simulation error between the RCS system simulation features and the reference feature, and optimize the network parameters based on the simulation error to improve the RCS system simulation accuracy.

[0154] This embodiment also provides a computer device, which is applicable to the case of an RCS system simulation test method based on a deep neural network, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the RCS system simulation test method based on a deep neural network proposed in the above embodiment.

[0155] This computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0156] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for simulating and testing an RCS system based on a deep neural network as proposed in the above embodiment.

[0157] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0158] Embodiment 3

[0159] This embodiment provides a method for simulating and testing an RCS system based on a deep neural network. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0160] To verify the effectiveness of the method for evaluating the RCS measurement quality based on a deep neural network of the present invention, a certain type of fighter model is selected as the test object for simulation experiments. The model includes multiple typical scattering points such as the nose, wings, and vertical tail. In the azimuth range of 0 - 360 degrees, RCS data is collected at 2-degree intervals. The experiments are respectively carried out under standard environment and interference environment, where the interference environment simulates the environmental interference in actual measurement by superimposing Gaussian white noise with a signal-to-noise ratio of 5 dB.

[0161] In the standard environment test, analyzing the RCS measurement results of the nose part (0-degree azimuth), the evaluation accuracy rate of the method of the present invention reaches 97.3%, while that of the traditional statistical method is 89.1%. In the complex scattering area at the connection between the wing and the fuselage (45-degree azimuth), the deviation between the evaluation result of the method of the present invention and the theoretical value remains within 3.2%, which is significantly better than the 8.7% deviation of the traditional method. When a measurement anomaly is artificially introduced at the 90-degree azimuth, the method of the present invention can detect and locate the problem within 0.5 seconds, and the measurement stability after the adaptive adjustment of the system parameters is improved by 85.4%.

[0162] In the test environment with noise interference introduced, analyzing the continuous measurement data in the azimuth range of 180 - 270 degrees, the evaluation result consistency of the method of the present invention reaches 94.2%, while that of the traditional method drops to 76.8%. Especially in the interval of 225 - 235 degrees where the target attitude changes rapidly, the parameter optimization strategy based on deep learning enables the signal processing parameters of the system to be adaptively adjusted, reducing the influence of measurement noise by 72.5%. Through the statistical analysis of 50 repeated tests, the standard deviation of the evaluation accuracy rate of the method of the present invention in a complex environment is 1.8%, verifying that the method has excellent stability and robustness.

[0163] The above experimental data fully demonstrate the significant advantages of the present invention in aspects such as measurement quality assessment, anomaly detection, and environmental adaptability, providing reliable technical support for the intelligent upgrade of the RCS test system.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A RCS system simulation test method based on deep neural network, characterized in that: include: Collecting RCS target echo characteristic data, segmenting the RCS target echo characteristic data according to a preset time window length to obtain a time series characteristic sequence; A dual-branch deep neural network structure is constructed based on the time series feature sequence, wherein the main branch uses a standard convolutional layer to process the time series feature sequence, and the auxiliary branch sets a spectrum reconstruction module to fuse the features extracted by the main branch and the auxiliary branch to obtain an RCS feature representation; A cross-scale feature correction unit is provided in the dual-branch deep neural network structure, and the cross-scale feature correction unit dynamically corrects the main and auxiliary branch features by calculating the mutual correlation coefficient between the main branch and the auxiliary branch feature graphs; Based on the output of the cross-scale feature correction unit, a high-frequency RCS simulation feature and a low-frequency RCS simulation feature are generated through a main branch decoder and an auxiliary branch decoder respectively, and a mutual correlation coefficient matrix is ​​used for adaptive fusion to obtain an RCS system simulation feature; The RCS feature representation is used as a reference feature, a simulation error between an RCS system simulation feature and the reference feature is calculated, and network parameters are optimized based on the simulation error to improve the RCS system simulation accuracy; The cross-scale feature correction unit includes a cross-correlation calculation module, a weight generation module and a feature correction module, wherein the cross-correlation calculation module is used to calculate the correlation matrix between two branch features, the weight generation module is used to generate the weight factor of the feature channel, and the feature correction module is used to optimize and correct the features; The correlation matrix is ​​calculated by extracting the main and auxiliary branch feature maps, adjusting them to the same feature space through projection transformation, and calculating the mutual correlation coefficient matrix , specifically including: The main branch feature map is extracted from each residual block. ,in , are the height and width of the feature map, respectively. is the number of feature channels, Indicates residual block level; The auxiliary branch feature map is obtained from the middle layer of the feature mapping network. ; Feature map of the main branch Rearrange and convert to matrix form ; The auxiliary branch features are dimensionally adjusted and normalized to be converted into a matrix form compatible with the main branch features. ; Calculate the correlation coefficient matrix between the main and auxiliary branch features : in, is the mutual correlation coefficient matrix, Indicates the direct correlation between features. Represents the autocorrelation matrix of the main branch features, The autocorrelation matrix representing the auxiliary branch characteristics; The weight generation module is a matrix of the mutual correlation coefficients. Perform row normalization to obtain the main branch feature weight factor ; Weight Factor Used to characterize the contribution of each feature channel of the main branch to the RCS feature representation; Obtain auxiliary branch feature weight factors through column normalization ; Weight Factor Used to characterize the contribution of each feature channel of the auxiliary branch to the RCS feature representation; The feature correction module converts the weight factor Applied to the main branch features to obtain the corrected feature map ; The weight factor Applied to the auxiliary branch features to obtain the corrected feature map .

2. The RCS system simulation test method based on deep neural network according to claim 1, characterized in that: Collecting the RCS target echo characteristic data, performing bandpass filtering on the target echo raw data to obtain a filtered echo signal, and performing complex sampling to obtain a discrete sampling sequence; The discrete sampling sequence is subjected to segmented windowing processing by applying a window function, and the windowed signal sequence is subjected to fast Fourier transform to obtain spectrum features, and amplitude spectrum and phase spectrum information are extracted from the spectrum features to form a time series feature sequence.

3. The RCS system simulation test method based on deep neural network according to claim 2, characterized in that: The main branch of the dual-branch deep neural network structure is used to process the time series feature sequence, and the secondary branch of the dual-branch deep neural network structure is used to process the frequency domain component, which is the amplitude spectrum and phase spectrum information; The main branch uses a standard convolutional layer to process the temporal feature sequence; And a residual connection is added every two convolutional layers to form a residual block structure; The residual block structure is a dual-path structure, including a main path and a shortcut connection path; The features processed by the main path are added to the features of the shortcut connection path at the element level, and the combined features are then activated by the ReLU function to obtain the final residual block output; Collecting feature maps output by each residual block, wherein the feature maps include shallow feature maps and deep feature maps; using an adaptive weight mechanism to perform feature fusion on the feature maps to obtain a fused feature representation; The fused feature representation is dimensionally compressed to obtain a main branch feature vector of fixed dimension. .

4. The RCS system simulation test method based on deep neural network according to claim 3, characterized in that: The secondary branch receives the amplitude spectrum and phase spectrum information, inputs the amplitude spectrum into a frequency domain feature extraction module composed of L layers of one-dimensional convolutional layers, and generates frequency distribution features; A compensation mechanism is introduced for the phase spectrum, based on the compensated phase spectrum and the original amplitude spectrum. Reconstructing frequency domain signals : in, is an imaginary unit, is the phase spectrum after compensation; The reconstructed frequency domain signal is converted into a sub-branch feature vector through a feature mapping network consisting of two fully connected layers. ; The main branch feature vector and the secondary branch eigenvector Splice on the feature dimension to form a fused feature vector; Perform dimension reduction mapping on the fused feature vector to obtain RCS feature representation .

5. The RCS system simulation test method based on deep neural network according to claim 4, characterized in that: Based on the corrected feature map Reconstruct it into high-frequency RCS simulation characteristics through the decoder; Based on the corrected feature map , reconstructed into low-frequency RCS simulation features through the auxiliary branch decoder; The high-frequency RCS simulation features and the low-frequency RCS simulation features are dimensionally unified and normalized to generate a high-frequency feature weight matrix and the low-frequency feature weight matrix ; The weight matrix is ​​multiplied element by element with the corresponding feature map, and the weighted high-frequency and low-frequency features are added element by element to obtain the RCS system simulation features. .

6. The RCS system simulation test method based on deep neural network according to claim 5, characterized in that: The feature reconstruction errors of the two are calculated based on the RCS system simulation features and the RCS feature representation, specifically including: Calculate the feature reconstruction error between the RCS system simulation features and the RCS feature representation ; For high frequency RCS characteristics, calculation and RCS characteristics representation The matching degree of mid-high frequency components; Evaluation and RCS feature representation of low frequency RCS features The consistency of mid- and low-frequency components; Analyze the correlation coefficient matrix during feature fusion changes, evaluate the fusion quality; The branch feature error is obtained by combining the matching degree of high-frequency RCS features, the consistency of low-frequency RCS features and the fusion quality. ; The total error is calculated based on the feature reconstruction error weight and the branch feature error weight, expressed as: in, is the feature reconstruction error weight, is the branch feature error weight.

7. The RCS system simulation test method based on deep neural network according to claim 6, characterized in that: Based on the changing trends of various errors, optimize parameter settings, including: If entering the main branch optimization, the residual block convolution layer parameters are optimized first, the adaptive weight mechanism parameters are adjusted, the main branch decoder parameters are updated, and the main feature extraction capability is enhanced; If the optimization of the main branch is completed, the feature mapping network parameters in the auxiliary branch are optimized, the auxiliary branch decoder parameters are updated, and the auxiliary feature extraction capability is improved; If the total error is lower than the preset threshold, or continuous The round error change is less than the convergence threshold, or the maximum number of iterations is reached , or the validation set error is continuous If there is no improvement in the first round, the optimization process is terminated and the network parameter optimization is completed.

8. A RCS system simulation test system based on a deep neural network, based on the RCS system simulation test method based on a deep neural network according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to collect RCS target echo characteristic data, and segment the RCS target echo characteristic data according to a preset time window length to obtain a time series characteristic sequence; The feature representation module builds a dual-branch deep neural network structure based on the time series feature sequence. The main branch uses a standard convolutional layer to process the time series feature sequence, and the auxiliary branch sets a spectrum reconstruction module to fuse the features extracted by the main branch and the auxiliary branch to obtain the RCS feature representation; The dynamic correction module sets a cross-scale feature correction unit in the dual-branch deep neural network structure, and dynamically corrects the main and auxiliary branch features by calculating the mutual correlation coefficient between the main branch and auxiliary branch feature maps; The simulation optimization module is used to generate high-frequency RCS simulation features and low-frequency RCS simulation features respectively, and perform adaptive fusion to obtain RCS system simulation features, use the RCS feature representation as a reference feature, calculate the simulation error between the RCS system simulation feature and the reference feature, optimize the network parameters based on the simulation error, and improve the RCS system simulation accuracy.

Images