CN-117591922-B - Ship target identification method and device based on RCS sequence and electronic equipment

CN117591922BCN 117591922 BCN117591922 BCN 117591922BCN-117591922-B

Abstract

The invention provides a ship target identification method, a device and electronic equipment based on an RCS sequence, wherein a transmitting device of a radar system transmits signals, and a receiving device receives feedback signals; and inputting the feedback signal into the trained improved OS-CNN model to obtain the object type of the returned feedback signal. The improved OS-CNN network is used in the radar target recognition field, has high classification recognition rate and has practical popularization value and application value. By introducing the advantages of offset convolution, attention mechanism and full-scale convolution, the improved OS-CNN network model can effectively extract the characteristics of target and interference RCS sequence data while keeping the lightweight and high efficiency, realizes accurate classification, and has important practical significance for improving the intelligent target recognition classification capability of the low-resolution radar.

Inventors

QUAN YINGHUI
Fan Hanxin
Lv Qinzhe
LIU LIYI
WU YAOJUN
ZHAO JIAQI

Assignees

西安电子科技大学

Dates

Publication Date: 20260512
Application Date: 20230925

Claims (7)

1. The ship target identification method based on the RCS sequence is characterized by comprising the following steps of: s100, transmitting signals through a transmitting device of a radar system, and receiving feedback signals through a receiving device; the feedback signals comprise echo signals returned by the real ship target and interference signals caused by the angle reflectors; s200, acquiring an improved OS-CNN model which is trained in advance, wherein the improved OS-CNN model adopts two branch channels and is realized by adding a channel attention module and a feature fusion module into the OS-CNN model; S300, inputting the feedback signal into the trained improved OS-CNN model, so that the improved OS-CNN model outputs the object type of the feedback signal; the training process of the improved OS-CNN model which is completed through pre-training comprises the following steps: S210, a static RCS database of a ship, an angular reflector and an angular array model is built in advance; S220, an OS-CNN model, a channel attention module and a feature fusion module are built, the network structure of the OS-CNN model is modified, and then the channel attention module and the feature fusion module are added into the OS-CNN model with the modified structure to obtain an improved OS-CNN model; S230, training the improved OS-CNN model by using the static RCS database to obtain a trained improved OS-CNN model; the modified OS-CNN model in S220 includes: The device comprises two branch channels, a feature fusion module and a full-connection layer, wherein the first branch channel consists of a channel attention module, the second branch channel consists of a plurality of custom convolution layers, a plurality of channel attention modules and a global average pooling layer, the back of each custom convolution layer is connected with one channel attention module, and the last channel attention module is connected with the global average pooling layer; s230 includes: s231, training a set and a test set respectively for a static RCS database; s232, extracting artificial features of the original data in the training set, and then inputting the extracted artificial features to a first branch channel and inputting the original data to a second branch channel; S233, applying attention weights to the artificial feature tensors through a channel attention module of the first branch channel to obtain weighted artificial feature vectors and obtaining depth feature vectors through the second branch channel; s234, carrying out feature fusion on the artificial feature tensor and the depth feature vector through the feature fusion module to obtain fusion features; S235, mapping the fusion characteristics into recognition classification results through a full connection layer; s236, adjusting parameters of the improved OS-CNN model according to the loss function of the identification and classification result, and returning to S232 until the iteration times are reached to obtain the trained improved OS-CNN model.
2. The RCS sequence-based ship target recognition method according to claim 1, wherein S210 comprises: s211, acquiring the sizes of a ship target, an angle reflector and an angle reflector array; s212, establishing three-dimensional models of each 1:1 of the three objects according to the size by utilizing SolidWorks three-dimensional drawing software; s213, importing the three-dimensional model into three-dimensional electromagnetic simulation software to obtain RCS amplitude characteristics of a ship target, an angle reflector array and a sea surface combined model within 360 degrees; S214, setting a frequency domain range and a plurality of frequency points in the frequency domain range, and acquiring an RCS amplitude value of the RCS amplitude characteristic at each frequency point; s215, carrying out normalization processing on the RCS amplitude, and longitudinally arranging normalized data to obtain a static RCS database.
3. The RCS sequence-based ship target recognition method according to claim 1, further comprising, after S236: And testing the trained improved OS-CNN model by using the test set.
4. The RCS sequence-based ship target recognition method of claim 1, wherein the applying the attention weight to the artificial feature tensor by the channel attention module of the first branch channel to obtain the weighted artificial feature vector comprises: a, carrying out pooling operation on an input feature x extracted from the artificial feature through self-adaptive average pooling to obtain an artificial feature tensor y with the shape of (batch_size, channels, 1); b, convolving the pooled artificial feature tensor y through a convolution layer of 1 multiplied by 1 to learn the attention weight of each channel; c, limiting the convolved output result between [0,1] through a Sigmoid activation function; And d, carrying out weighted multiplication on the input characteristic x according to the channel attention weight y to obtain a weighted artificial characteristic vector.
5. The RCS sequence-based ship target recognition method according to claim 1, wherein S300 comprises: S310, carrying out artificial feature extraction on the feedback signal, inputting signal features extracted by the artificial features into a first branch channel of the trained improved OS-CNN model, and inputting the feedback signal into a second branch channel; S320, processing the signal characteristics through a first branch channel to obtain a weighted artificial characteristic vector of the feedback signal, and processing the feedback signal through a second branch channel to obtain a depth characteristic vector of the feedback signal; S330, carrying out feature fusion on the depth feature vector and the weighted artificial feature vector of the feedback signal by the feature fusion module to obtain fusion features of the feedback signal; s340, classifying the fusion features through the fully-connected classification layer to obtain the object types of the returned feedback signals.
6. The utility model provides a naval vessel target identification device based on RCS sequence which characterized in that includes: the transmitting and receiving module transmits signals through a transmitting device of the radar system and receives feedback signals through a receiving device; the feedback signals comprise echo signals returned by the real ship target and interference signals caused by the angle reflectors; The system comprises an acquisition module, a feature fusion module and a channel attention module, wherein the acquisition module is used for acquiring an improved OS-CNN model which is trained in advance, and the improved OS-CNN model is realized by adding the channel attention module and the feature fusion module into the OS-CNN model; the classification module is used for inputting the feedback signal into the trained improved OS-CNN model so that the improved OS-CNN model outputs the object type of the feedback signal; the training process of the improved OS-CNN model which is completed through pre-training comprises the following steps: S210, a static RCS database of a ship, an angular reflector and an angular array model is built in advance; S220, an OS-CNN model, a channel attention module and a feature fusion module are built, the network structure of the OS-CNN model is modified, and then the channel attention module and the feature fusion module are added into the OS-CNN model with the modified structure to obtain an improved OS-CNN model; S230, training the improved OS-CNN model by using the static RCS database to obtain a trained improved OS-CNN model; the modified OS-CNN model in S220 includes: The device comprises two branch channels, a feature fusion module and a full-connection layer, wherein the first branch channel consists of a channel attention module, the second branch channel consists of a plurality of custom convolution layers, a plurality of channel attention modules and a global average pooling layer, the back of each custom convolution layer is connected with one channel attention module, and the last channel attention module is connected with the global average pooling layer; s230 includes: s231, training a set and a test set respectively for a static RCS database; s232, extracting artificial features of the original data in the training set, and then inputting the extracted artificial features to a first branch channel and inputting the original data to a second branch channel; S233, applying attention weights to the artificial feature tensors through a channel attention module of the first branch channel to obtain weighted artificial feature vectors and obtaining depth feature vectors through the second branch channel; s234, carrying out feature fusion on the artificial feature tensor and the depth feature vector through the feature fusion module to obtain fusion features; S235, mapping the fusion characteristics into recognition classification results through a full connection layer; s236, adjusting parameters of the improved OS-CNN model according to the loss function of the identification and classification result, and returning to S232 until the iteration times are reached to obtain the trained improved OS-CNN model.
7. The electronic equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus; a memory for storing a computer program; a processor for carrying out the method steps of any one of claims 1-5 when executing a program stored on a memory.

Description

Ship target identification method and device based on RCS sequence and electronic equipment Technical Field The invention belongs to the technical field of radar target identification, and particularly relates to a ship target identification method and device based on an RCS sequence and electronic equipment. Background The ocean space is wide, the resources are rich, the ocean space is an indispensable part of the country, the main right and the territory of the country are directly related, and continuous and thorough ocean monitoring is a basic support of the ocean country. The marine targets are various in variety, wide in distribution range and environment-changing in phantom, the targets are difficult to identify and distinguish, the detection difficulty is high, and the characteristics of the marine targets are important supports for realizing detection. In the actual sea target detection process, besides the echo of the target, the sea target detection method is also influenced by various interference factors such as foil strips, corner reflectors, active outboard interference and the like, so that the identification performance under the complex sea conditions is greatly influenced. Therefore, how to improve the problem of accurately identifying interference and targets by the radar has become a key problem to be solved. The radar corner reflector is a typical passive interference bait in electronic countermeasure, has the advantages of low manufacturing cost, obvious interference effect and the like, and can strongly reflect incident radar waves along an initial path, so that the radar corner reflector has a large scattering cross section and can interfere, deception and decoy a radar system. The RCS (radar cross section) scattering characteristics of the corner reflectors are analyzed and researched, so that theory and data support are provided for effectively identifying targets and corner decoys, and the method has important significance. Radar is the primary means of offshore object detection. Radar target characteristics mainly include radar cross-sectional area (Radar Cross Section, RCS), broadband characteristics, polarized scattering moment, and the like. At present, the identification method of ships and corner reflectors comprises Krogager polarization decomposition algorithm, micro Doppler characteristic-based method and the like. However, it is difficult for the radar to extract the polarization information of the target or interference, the complexity of the radar system needs to be increased to obtain the polarization information of the target, and the anti-interference practicability of the polarization processing mode is affected. The sea surface strong clutter can generate a hybrid spectrum in the echo signal spectrum, and the effect of the anti-interference method based on the micro Doppler characteristic can be influenced. Whereas RCS information is narrowband information that is available to almost all radars. The RCS sequence has the characteristics of small data size, good real-time performance and relatively simple processing technology. Compared with a high-resolution one-dimensional distance, the RCS has the advantage of being easier to acquire, and is an important basis of a target selection strategy in the conventional radar. The existing radar target recognition technology is mainly based on feature extraction of manual experience, the priori and actual effectiveness of the recognition process are low under the guidance of manual priori knowledge, and the recognition algorithm is only suitable for certain special cases and has poor generalization. Compared with traditional feature extraction, the development of deep learning technology provides a new direction for radar target recognition. Deep learning and neural networks are research hotspots in the field of artificial intelligence at present, and a large number of theoretical methods are included, but no general model can be applied to all data, and even for the same problem, if different parameters and super parameters are adopted, the results can be greatly different. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a ship target identification method and device based on an RCS sequence and electronic equipment. The technical problems to be solved by the invention are realized by the following technical scheme: in a first aspect, the present invention provides a ship target recognition method based on an RCS sequence, including: s100, transmitting signals through a transmitting device of a radar system, and receiving feedback signals through a receiving device; the feedback signals comprise echo signals returned by the real ship target and interference signals caused by the angle reflectors; s200, acquiring an improved OS-CNN model which is trained in advance, wherein the improved OS-CNN model adopts two branch channels and is realized by adding a channel attention mod