Sea surface target detection method based on data enhancement and recursive graph classification of transformer technology

Abstract

The application discloses a sea surface target detection method based on data enhancement and recursive graph classification of a Transform technology and belongs to the field of radar target detection. The sea surface target detection method comprises the following steps: obtaining a sea surface echo amplitude sequence by using a radar, pre-processing the sea surface echo amplitude sequence to obtain radar measured target echo amplitude data, and processing the radar measured target echo amplitude data into a data format suitable for being input into a Transform; generating historical item data and future item data of the measured target amplitude by using a Transform bidirectional prediction mechanism according to the pre-processed measured target amplitude data, and splicing the historical item data and the future item data with the measured target amplitude data to obtain enhanced target amplitude data; constructing the enhanced measured target amplitude data into a recursive graph data set; training the recursive graph data set by using a convolutional neural network to obtain a classifier model; and classifying a to-be-detected recursive graph by using the classifier model and outputting a classification result.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of radar target detection, and particularly relates to a sea surface target detection method based on data enhancement and recursive graph classification of a Transformer technology. BACKGROUND

[0002] Sea surface target detection is a key step in sea scene understanding and is widely used in fields such as marine environment monitoring, marine resource development, and navigation safety. However, the complex statistical characteristics of sea clutter and its uneven distribution in space and time bring great challenges to target detection. Sea clutter is affected by many factors such as sea state, weather, radar parameters, and operating frequency, and its intensity and distribution show high non-stationarity and non-Gaussianity. In particular, in a strong clutter environment, the detection and tracking of small targets have become a global technical problem. Researchers have proposed various detection methods to address this challenge. Traditional methods mainly include time-frequency representation, such as the method proposed by Victor C. Chen et al. in the article "Time-Frequency Transforms for Radar Imaging and Signal Analysis", which uses information in the time and frequency domains to analyze and process signals to adapt to changes in sea clutter. Deng Hua-wu et al. in the article "Unsupervised Segmentation of Synthetic Aperture Radar Sea Ice Images Based on Markov Random Field Model" use a specific filter to remove or suppress sea clutter and retain useful target signals. Although these traditional methods are effective under certain conditions, they often struggle to adapt to complex and changing sea conditions and data imbalance problems, especially in strong sea clutter, rapidly changing sea conditions, or when target samples are scarce, their performance and robustness may be significantly limited. SUMMARY

[0003] To overcome the deficiencies of the prior art, the purpose of the present application is to provide a sea surface target detection method based on data enhancement and recursive graph classification of a Transformer technology, which solves the problems in the prior art.

[0004] The purpose of the present application can be achieved by the following technical solutions:

[0005] The sea surface target detection method based on data enhancement and recursive graph classification of a Transformer technology comprises the following steps:

[0006] Obtain the sea surface echo amplitude sequence using a radar, preprocess it to obtain radar measured target echo amplitude data, and process it into a data format suitable for input into a Transformer;

[0007] Based on the preprocessed measured target amplitude data, the Transformer bidirectional prediction mechanism is used to generate historical and future data of the measured target amplitude, and then concatenated with the measured target amplitude data to obtain the enhanced target amplitude data.

[0008] The enhanced measured target amplitude data are used to construct a recursive graph dataset.

[0009] A classifier model is obtained by training a convolutional neural network on a recursive graph dataset; the classifier model is then used to classify the recursive graph under test and output the classification results.

[0010] The steps for preprocessing the sea surface echo amplitude sequence are as follows:

[0011] S11, extract the radar measured target data unit, add the real and imaginary parts and then calculate the absolute value to obtain the radar measured target amplitude data;

[0012] S12 defines the radar measured target amplitude data as an item in the Transformer system and sets up a virtual user to interact with the item.

[0013] Furthermore, the sea surface echo amplitude sequence is composed of N consecutive pulse echoes received by the radar, forming a time series dataset; each pulse echo data contains amplitude information.

[0014] Furthermore, the steps for generating historical and future data for the measured target amplitude include:

[0015] S21, by training backwards from right to left, the original input data is used to generate historical input data, thus generating historical item data. The expression is:

[0016] p(v k-1 =v k |S)

[0017] In the formula, Represents the original input sequence data set. v k Indicates the first term of the original input sequence data; v k-1 This represents the first historical item predicted based on the original input sequence data; this prediction process is repeated iteratively k times to obtain the set of historical item data. The superscript u indicates that the set is a set of sequence data that interacts with user u;

[0018] S22, utilizing The Transformer is fine-tuned using the input sequence, and the next term is predicted to generate future term data. The expression is as follows:

[0019]

[0020] In the formula, The predicted historical item data set is spliced with the original input sequence data set in the order of the historical item data set and the original input sequence data set to obtain an enhanced data set, that is, The tail item of is represented by The first future item data predicted according to the enhanced data set is represented by The historical item data set, the original input sequence data set and the future item data set are spliced together to obtain the balanced target amplitude data set

[0021] Further, the step of constructing the recurrence graph data set comprises:

[0022] S31, the balanced target amplitude data set is cut into several sub-segments with a length of N as input time series data, and for each data sub-segment x(t), an embedding vector set is constructed; the embedding vector construction expression is:

[0023] v(t) = {x(t), x(t+τ), x(t+2τ),..., x(t+(m-1)τ)}

[0024] In the formula, v(t) is a vector generated by embedding dimension and embedding delay, that is, an embedding vector, τ is a delay time used in the embedding process, and m is an embedding dimension;

[0025] S32, the distance between embedding vectors is represented by a recurrence matrix R; the element R i,j is defined as follows:

[0026] R i,j =||v(i)-v(j)||

[0027] Wherein, 1≤i,j≤n, n=N-(m-1)τ, and ||·|| represents Manhattan distance; then the recurrence matrix can be represented as:

[0028]

[0029] S33, the gray scale normalization processing of the recurrence matrix R is performed to obtain the recurrence graph data set; the formula of the gray scale normalization processing is:

[0030]

[0031] Wherein, G(i, j) is the gray value of the transformed image, R(i, j) represents the gray value of the original image; min and max represent the minimum and maximum gray values in the original image respectively.

[0032] Further, the specific steps of classifying the to-be-tested recursive graph by using the classifier model include:

[0033] S41, the recursive graph data set is transmitted into the first layer convolution layer of the convolution neural network, and the feature map of the recursive graph is output:

[0034]

[0035] Wherein f(·) is a nonlinear activation function; is the jth node of the convolution layer; M i is the range of the local receptive field; k is the convolution kernel; b c is the bias; the output of the convolution layer is the feature map of the recursive graph;

[0036] S42, the feature map generated by the convolution layer is input into the next layer pooling layer:

[0037] z i =f down (y1,y2,...y i ),y i ∈V

[0038] Wherein, f down (·) is a down-sampling function, and in the present application, the maximum pooling method is adopted, z i represents the output value of the ith node, and V represents the value range of the input value of the pooling layer;

[0039] S43, the output of the convolution layer and the pooling layer is input into the full connection layer, and the output of the full connection layer is the classification result, and the specific formula is as follows:

[0040]

[0041] Wherein, represents the output value of the ith node, that is, the number of feature map elements, w and b f are the weight matrix and the bias vector, f(·) represents a nonlinear activation function, represents the ith node of the current full connection layer, O f is the number of elements in the feature map output by the last layer.

[0042] The sea surface target detection system based on the data enhancement and recursive graph classification of the Transformer technology comprises:

[0043] A radar signal receiving module is used to obtain a sea surface echo sequence;

[0044] Data preprocessing module: used for preprocessing sea surface echo data to obtain radar measured sea surface echo amplitude data; according to the echo amplitude data obtained by preprocessing, the echo amplitude data is again processed into a data format suitable for being sent into the transformer;

[0045] Data generation module: generating historical item data and future item data of the measured target amplitude by using the bidirectional prediction mechanism of the transformer, and splicing the measured target amplitude data to obtain enhanced target amplitude data;

[0046] Recursive graph structure generation module: constructing the enhanced measured target amplitude data into a recursive graph data set;

[0047] Network model training module: used for training the recursive graph data set and obtaining a classifier model;

[0048] And the classification module: used for classifying the to-be-measured recursive graph by using the classifier model obtained by training, and outputting the result.

[0049] A computer storage medium stores a readable program, when the program runs, the above-mentioned sea surface target detection method based on data enhancement and recursive graph classification of transformer technology can be executed.

[0050] An electronic device, comprising: a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface complete the communication among each other through the communication bus;

[0051] The memory is used to store at least one executable instruction, and the executable instruction makes the processor execute the operation corresponding to the above-mentioned sea surface target detection method based on data enhancement and recursive graph classification of transformer technology.

[0052] A computer program product, comprising computer instructions, the computer instructions instruct a computing device to execute the operation corresponding to the above-mentioned sea surface target detection method based on data enhancement and recursive graph classification of transformer technology.

[0053] 1、The detection method of the application utilizes the data enhancement model of the transformer network, generates high-quality historical item and future item data through the bidirectional prediction mechanism, splices them to obtain enhanced data, constructs the enhanced data into a recursive graph data set, uses the recursive graph as a tool for quantitative analysis of recursion and nonlinearity to analyze the sea clutter data, and obtains a classifier model through the training of the convolutional neural network, so that the classifier can obtain good classification effect, improve the precision and effect of target detection.

[0054] 2、The application based on the Transformer to enhance data, compared with the existing data enhancement technology shows good performance. The traditional method usually needs to balance the data set, such as the ratio of target and non-target samples reaches 1:1 to obtain satisfactory results. And the application proposes a data enhancement method based on the Transformer technology, which can achieve excellent performance under the unbalanced 3:10 ratio, improve the efficiency and speed up the whole process. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0056] Figure 1 is the sea surface target detection flowchart of the application;

[0057] Figure 2 is the flowchart of the data enhancement based on the Transformer technology of the application;

[0058] Figure 3 is the performance comparison chart under different sampling methods of the application;

[0059] Figure 4 is the sea surface target detection system structure diagram of the application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0061] As shown in Figure 1 , the sea surface target detection method based on the data enhancement and recursive graph classification of the Transformer technology comprises the following steps:

[0062] S1, obtaining the sea surface echo amplitude sequence by radar, pre-processing the radar measured target echo amplitude data to obtain the radar measured target echo amplitude data, and processing into a data format suitable for inputting into the Transformer.

[0063] The sea surface echo amplitude sequence is composed of N consecutive pulse echoes received by the radar, forming a time series data set. Each pulse echo data contains its corresponding amplitude information; the specific steps of pre-processing the sea surface echo amplitude sequence are as follows:

[0064] S11, the radar measured target data unit is taken out, and the real part and the imaginary part are added and the absolute value is obtained to obtain the radar measured target amplitude data;

[0065] S12, the radar measured target amplitude data is defined as "items" in the Transformer system, and a virtual "user" is set to interact with these "items";

[0066] The number of items contained in each user in the Transformer system is the number of interactions; The Transformer system has a "timestamp" requirement, which is sequentially increased according to the arrangement order of the real radar data; Other data format requirements in the Transformer system are generated by a random function to generate simulated data.

[0067] S2, according to the preprocessed measured target amplitude data, the Transformer bidirectional prediction mechanism is used to generate historical item data and future item data of the measured target amplitude, and the historical item data, the measured target data and the future item data are spliced together to obtain enhanced target amplitude data;

[0068] The specific steps of obtaining the enhanced target amplitude data are:

[0069] S21, by means of right-to-left reverse training, the original input data is back- propagated to generate possible historical input data, that is, historical item data similar to the real input data is generated. For a given sequence S=[v k ,v k+1 ,...,v n ], where represents the kth to nth item in the original input data set S, which belongs to the item set The formula for Transformer to predict historical item data is as follows:

[0070]

[0071] In the formula, represents the original input sequence data set such as v k represents the first item of the original input sequence data; v k-1 represents the first historical item data predicted according to the original input sequence data. This prediction process is repeated k times, and the historical item data set is obtained, where the superscript u represents that the set is a sequence data set interacting with user u.

[0072] The formula is explained as follows: by using a given original input data set S=[v k ,vk+1 ,...,v n Predicting v k-1 ,in This represents the previous term in the sequence. First, the sequence is input into a Transformer to perform a history term generation operation, resulting in a history term v. k-1 Then, this history item v k-1 v with the original sequence input k to v n-1 The data is then recombined and used as input for the next operation. This process is repeated iteratively k-1 times, eventually generating data from v0 to v... k-1 The set of all k historical items is the historical item data set. This represents the (k-1)th term that interacts with user u.

[0073] Because an iterative prediction method is used, for the input sequence S = [v k ,v k+1 ,...,v n Each iteration of the prediction process passes these input sequences through the Transformer layer. This process facilitates the Transformer network model's learning and integration of contextual information from preceding elements, resulting in richer and more advanced temporal dynamic representations and contextual dependencies. Specifically, for each sequence in the original input set S, the first prediction process generates a history item v. k-1 At the same time, v k ,v k+1 ,...,v n After passing through the Transformer layer, the sequence is transformed into a new representation containing more comprehensive information. These new representations not only capture the internal relationships within the sequence but also consider the interactions between items. Therefore, the representation of each element is based on its own information and the cumulative information of all preceding elements, resulting in a more comprehensive set of historical item data. It more closely approximates the real data distribution. After multiple experiments, the number of iterations was set to 52, and the number of items in each input sequence set S was between 50 and 200. The Transformer network model showed significant predictive performance. This parameter configuration can produce high-quality prediction results.

[0074] S22, utilizing The input sequence is used to fine-tune the Transformer and predict the next term;

[0075] Bundle As input sequences, these are fed into the Transformer to fine-tune various parameters of the Transformer network model, enabling it to better understand and predict the dynamic changes in time series data. The principle behind this is described in S21. After fine-tuning, the parameters are then utilized... To make a prediction, we obtain the prediction term v. n+1 .

[0076] To continuously optimize the accuracy of predictions, this invention employs an iterative method for predicting future data: after the first prediction, the data is... The first term is truncated to obtain and v n+1 Fill to The last item was obtained This result is then fed back into the Transformer network model for prediction, and this step is repeated k-1 times. Here, the formula used to predict the next term is:

[0077]

[0078] In the formula, This refers to the augmented data set obtained by concatenating the predicted historical data set with the original input sequence data set in the order of the historical data set and the original input sequence data set. express The last item; Let represent the first future term data predicted based on the augmented dataset. This prediction process is repeated iteratively k times to obtain the future term dataset. We concatenate the historical data set, the original input sequence data set, and the future data set together to obtain the balanced target magnitude data set. Right now It includes historical data of target cells predicted by Transformer, measured target data, and future data of target cells predicted by Transformer.

[0079] S3, construct a recursive graph dataset from the equalized target amplitude data set;

[0080] The steps to construct a recursive graph dataset include:

[0081] S31, the enhanced measured target amplitude dataset The input time series data is divided into several segments of length N. For each data segment x(t), an embedding vector set is constructed. The expression for constructing the embedding vector is:

[0082] v(t) = {x(t), x(t+τ), x(t+2τ),...,x(t+(m-1)τ)}

[0083] wherein v(t) is a vector generated by embedding dimension and embedding time delay, τ is a delay time used in embedding process, and m is an embedding dimension;

[0084] S32, using a recurrence matrix R to represent the distance between embedding vectors; the element R i,j is defined as follows:

[0085] R i,j =||v(i)-v(j)||

[0086] wherein 1≤i,j≤n, n=N-(m-1)τ, and ||·|| represents Manhattan distance; then the recurrence matrix can be represented as:

[0087]

[0088] S33, performing gray scale normalization processing on the recurrence matrix R to obtain a recurrence graph data set; the formula for gray scale normalization processing is:

[0089]

[0090] wherein G(i,j) is the gray scale value of the transformed image, R(i,j) represents the gray scale value of the original image, min and max represent the minimum and maximum gray scale values in the original image, respectively. Through the above method, the recurrence graph data set of the sea surface measured data is finally obtained.

[0091] S4, training the recurrence graph data set by using a convolutional neural network to obtain a classifier model for classification, and finally obtaining a classification result to realize detection of sea surface targets.

[0092] The specific steps for training the recurrence graph data set by using a convolutional neural network to obtain a classifier model include:

[0093] 1) First, build a network model, and build a classic neural network model-LeNet-5, which includes a convolutional layer (C1), a normalization layer (N2), an activation layer (A3), a pooling layer (C4), a convolutional layer (C5), a normalization layer (N6), an activation layer (A7), a pooling layer (C8), a fully connected layer (F9), and an output layer (O10).

[0094] 2) Divide the obtained recurrence graph data set into two parts, a training set and a test set, wherein the ratio of the training set to the test set is 7:3.

[0095] 3) After the data segmentation is completed, a label is added to the recurrence graph, which is classified into two categories of target and clutter. The training set and the validation set are input into the network model, and after training, a classifier model is obtained.

[0096] Then the test set data is input into the trained classifier model; the classifier model processes the input test set data based on the network parameters obtained by its training.

[0097] The classifier model quickly outputs the corresponding classification result after internal operation; in this embodiment, the construction of the recurrence graph parameters is: time series length N = 1000, time delay τ = 11 and embedding dimension m = 5. The network parameters are: the number of C1 convolution kernels is 16, the number of C3 convolution kernels is 32, and the convolution kernel size is 5x5. The activation layer adopts the ReLU activation function. The pooling layer adopts the maximum pooling method, and the pooling matrix size is 2x2 and the step is 2.

[0098] The specific steps of classifying the recurrence graph data using the trained convolutional neural network include:

[0099] S41, the recurrence graph data set is input into the first layer of the convolutional neural network, and the feature map of the recurrence graph is output, and the specific formula is as follows:

[0100]

[0101] Where f(·) is a nonlinear activation function; is the jth node of the convolution layer; M i is the range of the local receptive field; k is the convolution kernel; b c is the bias; the output of the convolution layer is the feature map of the recurrence graph;

[0102] S42, the feature map generated by the convolution layer is input into the next layer of the pooling layer, and the specific formula is as follows:

[0103] z i =f down (y1,y2,...y i ),y i ∈V

[0104] Where f down (·) is a down-sampling function, and in this invention, the maximum pooling method is adopted, z i represents the output value of the ith node, and V represents the value range of the input value of the pooling layer.

[0105] S43, input the output of the convolutional layer and the pooling layer into the fully connected layer, and the output of the fully connected layer is the classification result. Specifically, the fully connected layer compresses the high-dimensional feature map into a low-dimensional representation, and each node of the last layer corresponds to the score or probability of a class, which realizes the end-to-end mapping from the original recursive graph data to the final classification result, and the specific formula is as follows:

[0106]

[0107] wherein, represents the output value of the i-th node, that is, the number of feature map elements, w and b f are the weight matrix and the bias vector, and f(·) represents a nonlinear activation function, represents the i-th node of the current fully connected layer, O f is the number of elements in the feature map output by the last layer. Generally, the first fully connected layer converts multiple feature maps into a vector, and the last fully connected layer acts as a softmax classification layer.

[0108] Embodiment 2

[0109] In this embodiment, the effect of the sea surface target detection method of the application is verified through a specific experiment.

[0110] In this embodiment, the data measured by Professor S. Haykin of McMaster University in Canada using IPIX radar in a real marine environment is used as the sea surface echo amplitude sequence obtained by the radar of the application. In the experiment, the method of the application, simple oversampling, Smote oversampling and random undersampling are used as comparative algorithms, the operation process of the method of the application and part of the data are as shown in Figure 2 , and the detection results of different algorithms under HH polarization are as shown in Figure 3 From the figure, it can be seen that the precision, recall and F1 coefficient of the method proposed in the application all maintain at a high level, among which the most important F1 coefficient can reach 85.5%, while the F1 coefficients of the simple oversampling, Smote oversampling and random undersampling methods are 79.8%, 80.54% and 73.01% respectively, so the data balancing method proposed in the application is superior to the other several classification methods in classification performance, and improves the efficiency and accelerates the whole processing process.

[0111] Embodiment 3

[0112] As shown in Figure 4 , in this embodiment, a sea surface target detection system based on data enhancement and recursive graph classification of the Transformer technology is proposed, which comprises:

[0113] A radar signal receiving module for obtaining a sea surface echo sequence;

[0114] Data preprocessing module: used for preprocessing the sea surface echo data to obtain radar measured sea surface echo amplitude data; according to the echo amplitude data obtained by preprocessing, the echo amplitude data is again processed into a data format suitable for being sent into the Transformer;

[0115] Data generation module: using the bidirectional prediction mechanism of the Transformer to generate historical item data and future item data of the measured target amplitude, and splicing the three to obtain enhanced target amplitude data;

[0116] Recursive graph structure generation module: receiving the preprocessed recursive matrix as input, and converting the matrix information into a recursive graph structure data set through a specific algorithm;

[0117] Network model training module: used for training the recursive graph data set and obtaining a classifier model;

[0118] Classification module: used for classifying the to-be-tested recursive graph by using the trained classifier model, and outputting the result.

[0119] The network model training module includes a data set cutting unit and a model training unit; the data set cutting unit is used for dividing the obtained recursive graph into a training set and a validation set, and the ratio of the three is 7:3. The model training unit is used for training and parameter optimization verification of the classifier model by using the obtained training set and validation set, and finally obtaining the classifier model.

[0120] The present application enhances the data based on the Transformer, and exhibits good performance compared with the existing data enhancement technology. The traditional method usually needs to balance the data set, such as the ratio of target and non-target samples reaching 1:1 to obtain satisfactory results. The present application proposes a data enhancement method based on the Transformer technology, and the classification result is more accurate and reliable under the unbalanced ratio of 3:10, and the classification performance is more stable, which improves the efficiency and accelerates the whole processing process.

[0121] The methods of the present application can be implemented in hardware, firmware, or software, or any combination thereof, and can be stored in or implemented with the aid of software or computer code stored in a recording medium as a computer program product without departing from the scope of the present application. The computer program product includes a computer readable medium, such as but not limited to the non-transitory machine-readable medium described above. The computer program product can also include packaging housing the computer-readable medium. The computer readable medium having stored thereon the software or computer code, which software or computer code, when executed by a computer, processor, or hardware, causes the computer, processor, or hardware to implement the methods described herein. The software or computer code can be stored in a computer readable medium, which is non-transitory, operable to store the software or computer code. Examples of the computer- readable medium include a hard disk, a CD-ROM, an optical disk, a DVD, a Blu-ray disk, a magnetic tape, a flash memory, a PROM, RAM, ROM, and the like. The computer readable medium can be portable, non-portable, or built-in (i.e., non-removable) with the computer or processor. The computer readable medium can be a recording medium, a memory device, or a combination thereof. The software or computer code can be stored in the computer readable medium, which is non-transitory, operable to store the software or computer code. Examples of the computer-readable medium include a hard disk, a CD-ROM, an optical disk, a DVD, a Blu-ray disk, a magnetic tape, a flash memory, a PROM, RAM, ROM, and the like. The computer readable medium can be portable, non-portable, or built-in (i.e., non-removable) with the computer or processor. The computer readable medium can be a recording medium, a memory device, or a combination thereof.

[0122] The foregoing is considered as illustrative only of the principles of the application. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the application to the exact construction and practice described. Accordingly, all such variations and modifications are intended to be included within the scope of the present application as defined in the following claims.

Claims

1. A sea surface target detection method based on Transformer technology for data augmentation and recursive graph classification, characterized in that, Includes the following steps: The radar obtains the sea surface echo amplitude sequence, preprocesses it to obtain the radar measured target echo amplitude data, and processes it into a data format suitable for input into Transformer. Based on the preprocessed measured target amplitude data, the Transformer bidirectional prediction mechanism is used to generate historical and future data of the measured target amplitude, and then concatenated with the measured target amplitude data to obtain the enhanced target amplitude data. The enhanced measured target amplitude data are used to construct a recursive graph dataset. A classifier model is obtained by training a recursive graph dataset using a convolutional neural network; The classifier model is used to classify the recurrence graph under test and the classification results are output. The steps for generating historical and future data for the measured target amplitude include: S21, by training backwards from right to left, the original input data is used to generate historical input data, thus generating historical item data. The expression is: In the formula, Represents the original input sequence data set. ; This represents the first item of the original input sequence data; This represents the first historical item predicted based on the original input sequence data; this prediction process is repeated iteratively. k Next, we obtain the historical data set. superscript u This indicates that the set is related to the user. u A collection of sequence data for interaction; S22, utilizing The Transformer is fine-tuned using the input sequence, and the next term is predicted to generate future term data. The expression is as follows: In the formula, This refers to the augmented data set obtained by concatenating the predicted historical data set with the original input sequence data set in the order of the historical data set and the original input sequence data set. ; express The last item; This represents the first future item predicted based on the augmented dataset; this prediction process is repeated iteratively. k Next, we obtain the future item data set. The historical data set, the original input sequence data set, and the future data set are concatenated to obtain the balanced target magnitude data set. .

2. The sea surface target detection method based on Transformer technology for data augmentation and recursive graph classification according to claim 1, characterized in that, The steps for preprocessing the sea surface echo amplitude sequence are as follows: S11, extract the radar measured target data unit, add the real and imaginary parts and then calculate the absolute value to obtain the radar measured target amplitude data; S12 defines the radar measured target amplitude data as an item in the Transformer system and sets up a virtual user to interact with the item.

3. The sea surface target detection method based on Transformer technology for data augmentation and recursive graph classification according to claim 1 or 2, characterized in that, The sea surface echo amplitude sequence consists of N consecutive pulse echoes received by the radar, forming a time series dataset; each pulse echo data contains amplitude information.

4. The sea surface target detection method based on Transformer technology for data augmentation and recursive graph classification according to claim 3, characterized in that, The steps for constructing the recursive graph dataset include: S31, the equalized target amplitude dataset is used as input time series data and divided into several segments of length N. For each data segment... Construct a set of embedded vectors; then the expression for constructing embedded vectors is: In the formula, An embedded vector is a vector generated by the embedding dimension and embedding delay. The delay time used in the embedding process, where m is the embedding dimension; S32, using a recursive matrix Represents the distance between embedding vectors; recursion matrix elements The definition is as follows: in, , , Let Manhattan distance be represented; then the recursive matrix is ​​represented as: S33, for the recursive matrix Perform grayscale normalization to obtain the recursive graph dataset; the formula for grayscale normalization is: in, The grayscale value of the transformed image. Represents the grayscale value of the original image; and These represent the minimum and maximum gray values ​​in the original image, respectively.

5. The sea surface target detection method based on Transformer technology for data augmentation and recursive graph classification according to claim 1, characterized in that, The specific steps for classifying the recurrence graph under test using a classifier model include: S41, feed the recursive graph dataset into the first convolutional layer of the convolutional neural network, and output the feature map of the recursive graph: in It is a non-linear activation function; The first convolutional layer One node; The extent of the local receptive field; For convolution kernel; The bias is used; the output of the convolutional layer is the feature map of the recursive graph. S42, input the feature map generated by the convolutional layer into the next pooling layer: in, The downsampling function uses max pooling. Indicates the first The output value of each node, Indicates the range of values ​​for the pooling layer input; S43, the outputs of the convolutional and pooling layers are fed into the fully connected layer. The output of the fully connected layer is the classification result, as shown in the following formula: in, Indicates the first i The output value of each node, i.e., the number of elements in the feature map. and These are the weight matrix and the bias vector. Represents a non-linear activation function. Indicates the current fully connected layer's... i 1 node It represents the number of elements in the feature map output from the last layer.

6. A sea surface target detection system based on Transformer technology for data augmentation and recursive graph classification, characterized in that, The system is used to implement the sea surface target detection method based on Transformer technology and recursive graph classification as described in any one of claims 1-5, the system comprising: Radar signal receiving module: used to obtain sea surface echo sequences; Data preprocessing module: used to preprocess sea surface echo data to obtain radar-measured sea surface echo amplitude data; based on the preprocessed echo amplitude data, it is further processed into a data format suitable for being sent into Transformer; Data generation module: Utilizes the Transformer bidirectional prediction mechanism to generate historical and future data of the measured target amplitude, and concatenates them with the measured target amplitude data to obtain the enhanced target amplitude data; Recursive graph structure generation module: Constructs a recursive graph dataset from the enhanced measured target amplitude data; Network model training module: used to train the recursive graph dataset and obtain the classifier model; And, the classification module: used to classify the test recursive graph using the trained classifier model and output the results.

7. A computer storage medium storing a readable program, characterized in that, When the program runs, it can perform the sea surface target detection method based on Transformer technology and recursive graph classification as described in any one of claims 1-5.

8. An electronic device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the sea surface target detection method based on Transformer technology and recursive graph classification as described in any one of claims 1-5.

9. A computer program product comprising computer instructions, characterized in that, The computer instructions instruct the computing device to perform the operations corresponding to the sea surface target detection method based on Transformer technology and recursive graph classification as described in any one of claims 1-5.

Images