A classification method, device and equipment for radar echo signals

Abstract

The present invention discloses a method, device and equipment for classifying radar echo signals, including: performing data preprocessing; selecting the nearest neighbor set when the autocorrelation degree of each sample in each pair of non-connectable constraint pairs is the largest as the corresponding set of same-class samples, thereby expanding the constraint pair set; generating a connection matrix from the expanded constraint pair set, calculating the objective function after each feature in the candidate feature subset is incorporated into the feature subset, selecting the feature index with the smallest objective function, updating the target feature subset, and generating training samples by performing feature selection on the original sample data set; using the training nearest neighbor classifier to classify unknown radar echo signals. The present invention considers that the nearest neighbor samples around each sample are different, and adaptively finds the most suitable nearest neighbor set for each sample to ensure that the connection between the elements is close, which is convenient for generating the optimal feature subset; considering the correlation between features and the local geometric structure of data, the accuracy of classifying radar echo signals is improved.

Description

Technical Field

[0001] The present invention relates to the field of pattern recognition, and particularly to a method, device, and equipment for classifying radar echo signals. Background Art

[0002] Radar echo signals are diverse. By analyzing radar echo signals, it can promote the development of military or civilian fields. Therefore, it is particularly important to accurately and efficiently analyze the characteristics of radar echo signals.

[0003] Currently, there are three existing feature selection methods: fully supervised feature selection method, semi-supervised feature selection method, and unsupervised feature selection method. Among them, the semi-supervised feature selection method has significant advantages in analyzing the characteristics of radar echo signals. Currently, for semi-supervised feature selection, a feature selection method based on constraint scoring has been proposed. However, this method does not consider the correlation between features, which may lead to redundancy in the selected feature subset. If the features selected in this way are used to classify radar echo signals, the accuracy rate will decrease, and misclassification will occur during classification.

[0004] As can be seen from the above, how to accurately and efficiently analyze the characteristics of radar echo signals is a problem to be solved currently. Summary of the Invention

[0005] The object of the present invention is to provide a method, device, and equipment for classifying radar echo signals, so as to solve the problem that the correlation between features is not considered in the current feature selection process, resulting in possible feature redundancy in the selected feature subset, and further causing misclassification when classifying radar echo signals.

[0006] To solve the above technical problems, the present invention provides a method for classifying radar echo signals, including: preprocessing known radar echo signals to form an original sample data set, where the original sample data set includes a labeled data set and an unlabeled data set, extracting the features of the original sample data set to form an initial feature set, and generating a set of connection constraints and a set of non-connection constraints using the labeled data set; initializing a target feature subset and a candidate feature subset, where the candidate feature subset is the difference between the initial feature set and the target feature subset, and setting the number of features to be selected and the maximum number of neighbors ; calculating the autocorrelation degree of each sample in each pair of non-connection constraints on the original sample data set for each of the neighbor sets of each sample, and respectively selecting the target neighbor set with the highest correlation degree with the original sample data set , where , the th neighbor set contains a sample with a nearest neighbor relationship, the th nearest neighbor set contains a sample with a nearest neighbor relationship, select the selected target nearest neighbor set as the same-class sample set, and generate an extended connection constraint pair set and a non-connection constraint pair set according to the same-class sample set, so as to generate an extended constraint pair set; according to the extended constraint pair set, generate two connection matrices respectively, construct a Laplacian matrix, calculate the objective function after each feature in the candidate feature subset is incorporated into the feature subset, select the feature index with the smallest objective function, add the selected feature to the target feature subset, and obtain the updated target feature subset and candidate feature subset; continuously iterate to continuously update the target feature subset and candidate feature subset until the established conditions are met, complete the iteration, stop the selection, and obtain the finally selected target feature subset; according to the finally selected target feature subset, perform feature selection on the original sample data set to generate training sample data, and use the nearest neighbor classifier trained with the training sample data to classify unknown radar echo signals.

[0007] Preferably, the calculation of each pair of non-connection constraint pairs the self-correlation degree of each sample in the nearest neighbor set on the original sample data set, and respectively select the target nearest neighbor set with the largest correlation degree with the original sample data set including:

[0008] For each pair of non-connection constraint pairs in the non-connection constraint pair set the steps are as follows: Step S11: Calculate the correlation degree between sets, let

[0009] and where, represents the sample set composed of the th nearest neighbor samples of represents an element thereof, represents the original sample data set; then the correlation degree between sets and is expressed as

[0010]

[0011] where, represents the function of taking the number of set elements;

[0012] Step S12: Calculate the autocorrelation degree of the set on the original sample dataset That is,

[0013]

[0014] where represents the autocorrelation degree of the set on the original sample dataset ; represents the correlation degree between the set and the set ; represents an element of ; represents the correlation degree between the set and the set ; represents an element on the original sample dataset except

[0015] Step S13: Find samples that have a neighbor relationship with and samples that have a neighbor relationship with , that is:

[0016]

[0017]

[0018] where represents the sample set composed of 's neighbor samples; represents the autocorrelation degree of the set on the original sample dataset ;

[0019] Respectively select the target neighbor sets with the largest correlation degree with the original sample dataset, where , the th neighbor set contains samples that have a neighbor relationship with , the th neighbor set contains samples that have a neighbor relationship with ;

[0020] Obtain using samples that have a neighbor relationship with and Set of similar samples , similarly obtain Set of similar samples .

[0021] Preferably, the set of extended connection constraint pairs and the set of non-connectable constraint pairs according to the set of similar samples include:

[0022] Set of extended mandatory connection constraint pairs and set of non-connectable constraint pairs :

[0023]

[0024]

[0025] Among them, the before the equation is the set of extended connection constraint pairs, and the after the equation is the set of connection constraint pairs before extension; the before the equation is the set of extended non-connection constraint pairs, and the after the equation is the set of non-connection constraint pairs before extension; is 's set of similar samples, is 's element; is 's set of similar samples, is 's element.

[0026] Preferably, the selection of the feature index with the smallest objective function includes:

[0027] When the number of features in the target feature subset is less than the number of features to be selected, that is, , perform steps S21 - S26:

[0028] Step S21: Generate two matrices and respectively from the set of extended mandatory connection constraint pairs and the set of non-connectable constraint pairs , and the specific elements in the matrix are constructed as follows:

[0029]

[0030]

[0031] Step S22: Calculate the diagonal matrix , the value of its diagonal elements is , and construct the Laplacian matrix ; Similarly, construct the Laplacian matrix ;

[0032] Step S23: Calculate the candidate feature subset for each feature incorporated into the feature subset and the resulting score:

[0033]

[0034] where is the function to calculate the sum of the diagonal elements of the matrix, represents the feature matrix on the feature subset ;

[0035] Step S24: Select the feature index with the minimum score :

[0036]

[0037] Step S25: Update the target feature subset by adding the feature index with the minimum score to the target feature subset i.e., ;

[0038] Step S26: Update the candidate feature subset i.e., ;

[0039] Repeat steps S21 - S26 until the number of features in the target feature subset is equal to the number of features to be selected, i.e., stop the selection.

[0040] Preferably, the preprocessing of the known radar echo signals to form the original sample dataset includes:

[0041] Collect the known radar echo signals and numerically form a dataset, perform feature representation on each data in the dataset, and perform normalization processing to obtain the original sample dataset, denoted as , , , represents the total number of data samples in the dataset, where represents the th sample data, represents the number of features of the sample data;

[0042] The feature matrix of the sample data is represented as where is the ​​A feature vector, and it is normalized to the interval [0, 1].

[0043] Preferably, the preprocessing of the known radar echo signal further includes: selecting data from the data set according to the known radar echo signal, adding labels to form a labeled data set; letting the selected labeled data set be and , respectively representing the radar echo signal sets when there are free electrons and when there are no free electrons; generating a must-connect constraint pair set and a cannot-connect constraint pair set , that is

[0044]

[0045]

[0046] Preferably, the initialization of the target feature subset and the candidate feature subset includes:

[0047] Initializing the target feature subset , all feature sets are , the candidate feature subset is ; and setting the maximum number of nearest neighbors required and the number of selected features .

[0048] Preferably, the classification of the unknown radar echo signal by using the nearest neighbor classifier trained with the training sample data includes:

[0049] Performing feature selection on the original sample data set , where according to the finally selected target feature subset to obtain the training sample data set ;

[0050] Using the nearest neighbor classifier, classifying the unknown radar echo signal with the samples in the training sample data set as the training data. The classification result of 1 indicates that there are free electrons in the radar echo signal, and the classification result of 2 indicates that there are no free electrons in the radar echo signal.

[0051] The present invention also provides a classification device for radar echo signals, including:

[0052] A data processing module, which preprocesses the known radar echo signal to form an original sample data set. The original sample data set includes a labeled data set and an unlabeled data set, extracts the features of the original sample data set to form an initial feature set, and generates a connection constraint pair set and a non-connection constraint pair set by using the labeled data set;

[0053] A setting module, which initializes a target feature subset, calculates a candidate feature subset, where the candidate feature subset is the difference between the initial feature set and the target feature subset, and sets the number of features to be selected;

[0054] An expansion module, which sets a maximum neighbor value , calculates the autocorrelation degree of each sample in each pair of non-connectable constraint pairs on the original sample data set, and respectively selects the target neighbor set with the highest correlation degree with the original sample data set , where , the th neighbor set contains samples having a neighbor relationship with , the th neighbor set contains samples having a neighbor relationship with

[0055] The selected target neighbor set is used as the set of similar samples, and the set of connection constraints and the set of non-connectable constraints are expanded according to the set of similar samples, so as to generate an expanded set of constraint pairs;

[0055] A selection module, according to the expanded set of constraint pairs, respectively generates two connection matrices, constructs a Laplacian matrix, calculates the objective function after each feature in the candidate feature subset is incorporated into the feature subset, selects the feature index with the smallest objective function, and adds the selected feature to the target feature subset to obtain an updated target feature subset and a candidate feature subset;

[0056] An iteration module continuously iterates to update the target feature subset and the candidate feature subset until a predetermined condition is met, then completes the iteration and stops the selection to obtain the finally selected target feature subset;

[0057] A classification module, according to the finally selected target feature subset, performs feature selection on the original sample data set to generate training sample data, and uses the nearest neighbor classifier trained with the training sample data to classify unknown radar echo signals.

[0058] This invention also provides a classification device for radar echo signals, including:

[0059] A memory, which is used to store a computer program;

[0060] A processor, which is used to implement the steps of the classification method for a radar echo signal as described above when executing the computer program.

[0061] In the classification method of radar echo signals provided by the present invention, during the feature selection process, since the situations of the neighboring samples around each sample are different, the adaptive neighbor iterative constrained scoring feature selection method is adopted to find the most suitable neighbor set for each sample, ensuring that the elements in the found neighbor set are closely related, thereby making the relationship between features stronger and facilitating the generation of an optimal feature subset. In addition, the relevance between features and the local geometric structure of the data are considered to make up for the redundancy phenomenon in the selected feature subset, improving the accuracy of the classification accuracy of radar echo signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 It is a flowchart of the first specific embodiment of the classification method of radar echo signals provided by the present invention;

[0064] Figure 2 It is a flowchart of the second specific embodiment of the classification method of radar echo signals provided by the present invention;

[0065] Figure 3 It is a relationship diagram between the number of feature subsets selected by the present invention and the constraint scoring and the classification accuracy rate of radar echo signals;

[0066] Figure 4 It is a structural block diagram of a classification device for radar echo signals provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The core of the present invention is to provide a classification method, device and equipment for radar echo signals.

[0068] In order to enable those skilled in the art to better understand the solution of the present invention, the following will further elaborate on the present invention in conjunction with the drawings and specific embodiments. Obviously, the described embodiments are only 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 belong to the scope of protection of the present invention.

[0069] Embodiment 1

[0070] Please refer to Figure 1 , Figure 1Flow chart of the first specific embodiment of the radar echo signal classification method provided by the present invention; the specific operation steps are as follows:

[0071] Step S101: Collect known radar echo signals and digitize them to form a data set. Perform feature representation on each data in the data set and perform normalization processing to obtain the original sample data set, denoted as , , , , where represents the th sample data,

[0072] The feature matrix of the sample data is expressed as , where is the th feature vector and is normalized to the interval [0, 1];

[0073] According to the known radar echo signals, select data in the data set and add labels to form a labeled data set , and the remaining is an unlabeled data set , that is , , where represents the labeled data set, represents the unlabeled data set, represents the number of labeled data, represents the number of unlabeled data, represents the total number of data samples in the data set; according to the labeled radar echo signal set generate a must - connect constraint pair set and a cannot - connect constraint pair set . A pair of radar echo signals in the must - connect constraint pair set belongs to the same category, while a pair of radar echo signals in the cannot - connect constraint pair set belongs to different categories, that is

[0074]

[0075]

[0076] Step S102: Initialize the target feature subset , all feature sets are , and the candidate feature subset is​ ; and set the maximum number of nearest neighbors required and the number of selected features ;

[0077] Step S103: Expand the set of must - connect constraint pairs and the set of cannot - connect constraint pairs . For each pair of cannot - connect constraint pairs in the set of cannot - connect constraint pairs , the following steps are included:

[0078] (1) Calculate the correlation degree between sets. Let and , where represents the sample set composed of nearest - neighbor samples of , represents an element among them, , represents the original sample data set; then the correlation degree between the sets and is expressed as

[0079]

[0080] where represents the function of taking the number of set elements;

[0081] (2) According to the above formula, calculate the autocorrelation degree of the set on the original sample data set , that is

[0082]

[0083] where represents the autocorrelation degree of the set on the original sample data set , represents the correlation degree between the set and the set , represents an element on represents the correlation degree between the set and the set , represents an element on the original sample data set excluding ;

[0084] (3) Find samples that have a nearest - neighbor relationship with and samples that have a nearest - neighbor relationship with​ Samples with a neighbor relationship, that is:

[0085]

[0086]

[0087] Among them, represents a sample set composed of 's nearest neighbor samples, represents the autocorrelation degree of the set on the original sample data set ;

[0088] Respectively select the target nearest neighbor set with the highest correlation with the original sample data set, where the th nearest neighbor set contains samples with a neighbor relationship with the th nearest neighbor set contains samples with a neighbor relationship with

[0089] (4)Use samples and with a neighbor relationship to obtain 's set of similar samples , and similarly obtain 's set of similar samples ;

[0090] (5)Expand the set of must-connect constraint pairs and the set of cannot-connect constraint pairs :

[0091]

[0092]

[0093] Among them, the before the equation is the expanded set of connect constraint pairs, and the after the equation is the set of connect constraint pairs before expansion; the before the equation is the expanded set of non-connect constraint pairs, and the after the equation is the set of non-connect constraint pairs before expansion; is 's set of similar samples, is 's element; is a set of similar samples is an element in

[0094] Step S104: When the number of features in the target feature subset is less than the number of features to be selected, that is :

[0095] (1) Generate two matrices and from the extended must-link constraint pair set and respectively. The specific elements in the matrices are constructed as follows:

[0096]

[0097]

[0098] (2) Calculate the diagonal matrix , the value of its diagonal elements is , and construct the Laplacian matrix ; similarly, construct the Laplacian matrix ;

[0099] (3) Calculate the score of each feature in the candidate feature subset after being incorporated into the feature subset :

[0100]

[0101] where is a function to calculate the sum of the diagonal elements of the matrix, represents the feature matrix on the feature subset ;

[0102] (4) Select the feature index with the minimum score:

[0103]

[0104] (5) Update the target feature subset by adding the feature index with the minimum score to the target feature subset , that is ;

[0105] (6) Update the candidate feature subset , that is ;

[0106] Step S105: Repeat Step S104 until the number of features in the target feature subset is equal to the number of features to be selected, that is Stop selection when

[0107] For the original sample dataset Perform feature selection to form a training sample dataset, denoted as , where .

[0108] Step S106: Collect unknown radar echo signals to form an echo signal sample . According to the target feature subset selected by the above feature selection, perform feature selection on the sample data to generate a reduced sample data . Use the nearest neighbor classifier, with the samples in the training sample dataset as training data to classify unknown radar echo signals.

[0109] Embodiment 2

[0110] Based on the above embodiments, in this embodiment, the UCI dataset Ionosphere is selected as the training data, and this dataset is an ionospheric radar echo dataset. The present invention extracts and analyzes the characteristics of lightning echo signals, and then classifies the ionospheric radar echo signals, and judges whether there are free electrons in the ionosphere according to the classification results of the radar echoes. The dataset contains a total of 351 data samples, and each sample contains 34 attributes. Each data sample represents a radar echo, each radar echo has 17 pulse numbers, and each pulse number has two electromagnetic signals. According to a series of analyses, the attribute values of each sample are numericalized. The specific implementation steps are as follows:

[0111] Step S201: Data preprocessing

[0112] Collect known radar echo signals and numerically form a dataset, perform feature representation on each data in the dataset, and perform normalization processing to obtain the original sample dataset, denoted as , , , represents the total number of data samples in the dataset, where represents the th sample data, represents the number of features of the sample data;

[0113] The feature matrix of the sample data is represented as , where is the th feature vector and is normalized to the interval [0, 1]; ​​

[0114] Select data from the dataset according to the known radar echo signal, and add labels to form a labeled dataset; let the selected labeled dataset be and , representing the radar echo signal sets when there are free electrons and no free electrons respectively; generate a set of must-link constraint pairs and a set of cannot-link constraint pairs , that is

[0115]

[0116]

[0117] In this embodiment, set the total number of samples , the number of features , and select labeled data in each class.

[0118] Step S202: Feature selection

[0119] 1. Initialize the target feature subset , the set of all features is , and the candidate feature subset is ; and set the maximum number of nearest neighbors required = 10 and the number of selected features .

[0120] 2. Expand the set of must-link constraint pairs and the set of cannot-link constraint pairs . For each pair of cannot-link constraint pairs in the set of cannot-link constraint pairs , the following steps are included:

[0121] (1) Find samples that have a nearest neighbor relationship with and samples that have a nearest neighbor relationship with , that is:

[0122]

[0123]

[0124] Among them, represents the sample set composed of 's nearest neighbor samples, represents the set in the original sample dataset The autocorrelation degree on;

[0125] (a) Calculate the correlation degree between sets. Let and , where represents the sample set composed of 's nearest neighbor samples, represents an element among them, , represents the original sample data set; then the correlation degree between sets and is expressed as

[0126]

[0127] where represents the function of taking the number of set elements;

[0128] (b) According to the above formula, calculate the autocorrelation degree of the set on the original sample data set , that is

[0129]

[0130] where represents the autocorrelation degree of the set on the original sample data set , represents the correlation degree between the set and the set , represents 's one element, represents the correlation degree between the set and the set , represents except the original sample data set 's one element;

[0131] (c) Respectively select the target nearest neighbor set with the largest correlation degree with the original sample data set, where , the th nearest neighbor set contains samples with a nearest neighbor relationship with , the th nearest neighbor set contains samples with a nearest neighbor relationship with ;

[0132] (2) Use ones and The set of similar samples obtained from samples with a neighbor relationship is ; similarly, the set of similar samples obtained from samples with a neighbor relationship is ;

[0133] (3) Expand the must - connect constraint pair set and the cannot - connect constraint pair set :

[0134]

[0135]

[0136] Among them, the before the equation is the expanded must - connect constraint pair set, and the after the equation is the must - connect constraint pair set before expansion; the before the equation is the expanded cannot - connect constraint pair set, and the after the equation is the cannot - connect constraint pair set before expansion; is the set of similar samples, is an element in; is the set of similar samples, is an element in.

[0137] 3. When the number of features in the target feature subset is less than the number of features to be selected, that is :

[0138] (1) Generate two matrices and respectively from the expanded must - connect constraint pair set and the cannot - connect constraint pair set . The specific elements in the matrices are constructed as follows:

[0139]

[0140]

[0141] (2) Calculate the diagonal matrix , the values of its diagonal elements are , and construct the Laplacian matrix ; similarly, construct the Laplacian matrix ;

[0142] (3) Calculate each feature in the candidate feature subset Incorporated into the feature subset The score after:

[0143]

[0144] where is a function to calculate the sum of the diagonal elements of a matrix, represents the feature matrix on the feature subset ;

[0145] (4) Select the feature index with the minimum score :

[0146]

[0147] (5) Update the target feature subset by adding the feature index with the minimum score to the target feature subset i.e., ;

[0148] (6) Update the candidate feature subset i.e., ;

[0149] 4. Repeat step 3 until the number of features in the target feature subset is equal to the number of features to be selected, i.e., stop selecting;

[0150] 5. Perform feature selection on the original sample data set to form a training sample data set, denoted as , where .

[0151] Step S203: The present invention conducts experiments on the Ionosphere data set. According to the target feature subset obtained from the above feature selection template, perform feature selection on the sample data to generate a reduced sample data . Use the nearest neighbor classifier, with as the training data to classify the radar echo signal. The classification result of 1 represents the radar echo signal when there are free electrons in the ionosphere, and the classification result of 2 represents the radar echo signal when there are no free electrons in the ionosphere.

[0152] In this embodiment, the number of new users is set to 35, that is, the number of unknown radar echo signals is 35, and the radar echo signal analysis module is repeated 100 times. Adopt the 10-fold cross-validation method, and take the average value of the ten rounds of experiments as the experimental result, as Figure 2 shown. Summarize Figure 2 of the curve, Table 1 gives the highest accuracy and standard deviation of the present invention and the feature selection method based on constraint scoring. From Figure 3As can be seen from Table 1, compared with the constrained scoring, the present invention can consider the correlation between features under the same features and obtain higher classification accuracy.

[0153] Table 1 Comparison of recognition results between the method of the present invention and the constrained scoring method

[0154]

[0155] Example 3

[0156] Please refer to Figure 3 , Figure 3 which is a structural block diagram of a classification device for radar echo signals provided by an embodiment of the present invention; the specific device may include:

[0157] A data processing module preprocesses known radar echo signals to form an original sample data set. The original sample data set includes a labeled data set and an unlabeled data set, extracts features of the original sample data set to form an initial feature set, and generates a connection constraint pair set and a non-connection constraint pair set by using the labeled data set;

[0158] A setting module initializes a target feature subset, calculates a candidate feature subset, where the candidate feature subset is the difference between the initial feature set and the target feature subset, and sets the number of features to be selected;

[0159] An expansion module sets a maximum neighbor value , calculates the autocorrelation degree of each sample in each pair of non-connection constraint pairs on the original sample data set for each of the neighbor sets of each sample, respectively selects a target neighbor set with the largest correlation degree with the original sample data set , where , the th neighbor set contains samples having a neighbor relationship with , the th neighbor set contains samples having a neighbor relationship with , and uses the selected target neighbor set as a set of like samples, and expands the connection constraint pair set and the non-connection constraint pair set according to the set of like samples, thereby generating an expanded constraint pair set;

[0160] A selection module, according to the expanded constraint pair set, respectively generate two connection matrices, construct a Laplacian matrix, calculate the objective function after incorporating each feature in the candidate feature subset into the feature subset, select the feature index with the minimum objective function, and add the selected feature to the target feature subset to obtain the updated target feature subset and candidate feature subset;

[0161] An iteration module, continuously iterate to update the target feature subset and candidate feature subset until the established conditions are met, then complete the iteration and stop the selection to obtain the finally selected target feature subset;

[0162] A classification module, according to the finally selected target feature subset, perform feature selection on the original sample data set to generate training sample data, and use the nearest neighbor classifier trained with the training sample data to classify the unknown radar echo signal.

[0163] The device of this embodiment is used to implement the aforementioned classification method of radar echo signals. Therefore, the specific implementation manners in the classification device of radar echo signals can be seen in the embodiment part of the classification method of radar echo signals in the previous text. For example, the data processing module 100, the setting module 200, the expansion module 300, the selection module 400, the iteration module 500, and the classification module 600 are respectively used to implement steps S101, S102, S103, S104, S105, and S106 in the above-mentioned classification method of radar echo signals. Therefore, their specific implementation manners can refer to the descriptions of the corresponding respective part embodiments and will not be elaborated here.

[0164] Embodiment 4

[0165] The specific embodiment of the present invention also provides a classification device for radar echo signals, including: a memory for storing a computer program; a processor for implementing the steps of the above-mentioned classification method of a radar echo signal when executing the computer program.

[0166] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method part.

[0167] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0168] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0169] The classification method and device of the radar echo signal provided by the present invention have been introduced in detail above. Specific examples are used herein to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art in the technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A classification method for radar echo signals, characterized in that, Including: Preprocess the known radar echo signals to form an original sample data set, where the original sample data set includes a labeled data set and an unlabeled data set, extract the features of the original sample data set to form an initial feature set, and generate a connection constraint pair set and a non - connection constraint pair set using the labeled data set; Initialize the target feature subset and the candidate feature subset. The candidate feature subset is the difference between the initial feature set and the target feature subset, and set the number of features to be selected and the maximum number of nearest neighbors ; Calculate the self-correlation degree of each sample in each pair of non-connectable constraint pairs on the original sample dataset, and select the target nearest neighbor set with the highest correlation with the original sample dataset respectively. Among them , the th nearest neighbor set contains samples that have a nearest neighbor relationship with , and the th nearest neighbor set contains samples that have a nearest neighbor relationship with . Take the selected target nearest neighbor set as the set of similar samples, and expand the connectable constraint pair set and the non-connectable constraint pair set according to the set of similar samples, so as to generate an expanded constraint pair set; According to the expanded constraint pair set, generate two connection matrices respectively, construct a Laplacian matrix, calculate the objective function after each feature in the candidate feature subset is incorporated into the feature subset, select the feature index with the minimum objective function, and add the selected feature to the target feature subset to obtain the updated target feature subset and candidate feature subset; Continuously iterate to update the target feature subset and the candidate feature subset until the established conditions are met, then complete the iteration and stop the selection to obtain the finally selected target feature subset; According to the finally selected target feature subset, perform feature selection on the original sample data set to generate training sample data, and use the nearest - neighbor classifier trained with the training sample data to classify the unknown radar echo signals.

2. The method according to claim 1, wherein Calculating the self-correlation degree of each sample in each pair of non-connectable constraint pairs in the nearest neighbor set of each sample in the original sample dataset, and respectively selecting the target nearest neighbor set with the highest correlation degree with the original sample dataset including: Pairs of non-connectable constraint pairs For each pair of non-connectable constraint pairs , the following steps are included: Step S11: Calculate the relevance between sets, and let and , where represents the sample set composed of 's nearest neighbor samples, represents one of its elements, , represents the original sample data set; then the relevance between sets and is expressed as ； Among them, A function representing the number of elements in a set; Step S12: According to the above formula, calculate the autocorrelation degree of the set in the original sample data set , that is ； Among them, represents a set of the autocorrelation degree on the original sample data set . represents a set and a set of the correlation degree between them represents an element of represents a set and a set of the correlation degree between them represents an element on the original sample data set excluding . Step S13: Find a number of samples with a neighbor relationship and a number of samples with a neighbor relationship, i.e.: ； ； Among them, represents a sample set composed of 's nearest neighbor samples, represents the autocorrelation degree of the set on the original sample data set ; Select the target nearest neighbor sets with the highest relevance to the original sample dataset respectively , where , the th nearest neighbor set contains samples with a nearest neighbor relationship with , and the th nearest neighbor set contains samples with a nearest neighbor relationship with ; Utilize and obtain the set of similar samples with neighboring samples, and similarly obtain the set of similar samples and .

3. The method according to claim 2, characterized in that, The expanding the connection constraint pair set and the non - connection constraint pair set according to the same - class sample set includes: Expansion must connect to a set of constraint pairs and cannot connect to a set of constraint pairs : ； ； Among them, before the equation is the set of extended connection constraint pairs, and after the equation is the set of connection constraint pairs before extension; before the equation is the set of non-connection constraint pairs after extension, and after the equation is the set of non-connection constraint pairs before extension; is the set of similar sample sets, is an element in; is the set of similar sample sets, is an element in.

4. The method according to claim 3, characterized in that, The selecting the feature index with the minimum objective function includes: When the number of features in the target feature subset is less than the number of features to be selected, that is perform steps S21 - S26: Step S21: Generate two matrices respectively from the expanded set of must - connect constraint pairs and the set of cannot - connect constraint pairs The specific elements in the matrices are constructed as follows: and ​ ； ； Step S22: Calculate the diagonal matrix , the values of its diagonal elements are , and construct the Laplacian matrix ; Similarly, construct the Laplacian matrix ; Step S23: Calculate the candidate feature subset for each feature incorporated into the feature subset and the resulting score: ； Among them is a function for calculating the sum of the diagonal elements of a matrix, represents the feature matrix on the feature subset ; Step S24: Select the feature index with the minimum score : ； Step S25: Update the target feature subset by adding the feature index with the minimum score to the target feature subset among them, that is ; Step S26: Update the candidate feature subset , that is ; Repeat steps S21 - S25 until the number of features in the target feature subset is equal to the number of features to be selected, i.e., stop selection when this occurs.

5. The method according to claim 4, wherein The preprocessing the known radar echo signals to form an original sample data set includes: Collect known radar echo signals and numerically form a data set. Represent the features of each data in the data set and perform normalization processing to obtain the original sample data set, denoted as , , , denotes the total number of data samples in the data set, where denotes the th sample data, denotes the number of features of the sample data; The feature matrix of the sample data is represented as , where is the th feature vector and is normalized to the interval [0, 1].​​ 6. The method according to claim 5, characterized in that, The preprocessing of the known radar echo signals further includes: selecting data from the dataset and adding labels to form a labeled dataset; letting the selected labeled dataset be and , , respectively representing the radar echo signal sets when there are free electrons and when there are no free electrons; generating a must - connect constraint pair set and a cannot - connect constraint pair set , that is ； 。 7. The method according to claim 6, characterized in that, The initializing the target feature subset and the candidate feature subset includes: Initialize the target feature subset , all feature sets are , the candidate feature subset is ; and set the required maximum number of nearest neighbors and the number of selected features , where .

8. The method according to claim 1, wherein The classifying the unknown radar echo signals using the nearest - neighbor classifier trained with the training sample data includes: Based on the finally selected target feature subset, for the original sample dataset , where feature selection is performed to obtain the training sample dataset ; Using the nearest neighbor classifier, with the samples in the training sample dataset as the training data to classify unknown radar echo signals. A classification result of 1 indicates that there are free electrons in the radar echo signal, and a classification result of 2 indicates that there are no free electrons in the radar echo signal.

9. A classification device for radar echo signals, characterized in that, Including: A data processing module that preprocesses the known radar echo signals to form an original sample data set, where the original sample data set includes a labeled data set and an unlabeled data set, extracts the features of the original sample data set to form an initial feature set, and generates a connection constraint pair set and a non - connection constraint pair set using the labeled data set; A setting module that initializes the target feature subset, calculates the candidate feature subset, where the candidate feature subset is the difference between the initial feature set and the target feature subset, and sets the number of features to be selected; Expansion module, set the maximum number of nearest neighbors , calculate each pair of non-connectable constraint pairs for each sample in the self-correlation degree of the nearest neighbor sets on the original sample data set, and select the target nearest neighbor set with the largest correlation degree with the original sample data set , where , the th nearest neighbor set contains samples with a nearest neighbor relationship with , the th nearest neighbor set contains samples with a nearest neighbor relationship with , and use the selected target nearest neighbor set as the set of similar samples, and expand the connectable constraint pair set and non-connectable constraint pair set according to the set of similar samples, so as to generate an expanded constraint pair set; A selection module that, according to the expanded constraint pair set, generates two connection matrices respectively, constructs a Laplacian matrix, calculates the objective function after each feature in the candidate feature subset is incorporated into the feature subset, selects the feature index with the minimum objective function, and adds the selected feature to the target feature subset to obtain the updated target feature subset and candidate feature subset; An iteration module that continuously iterates to update the target feature subset and the candidate feature subset until the established conditions are met, then completes the iteration and stops the selection to obtain the finally selected target feature subset; A classification module that, according to the finally selected target feature subset, performs feature selection on the original sample data set to generate training sample data, and uses the nearest - neighbor classifier trained with the training sample data to classify the unknown radar echo signals.

10. A classification device for radar echo signals, characterized in that, Including: A memory for storing a computer program; A processor for implementing the steps of a classification method for a radar echo signal as described in any one of claims 1 to 8 when executing the computer program.

Images