A Radar Multiple Echo Binary Classification Method Based on EasyEnsemble Algorithm
By merging multiple echoes into one class using the EasyEnsemble algorithm and designing a binary classifier, the problem of sample imbalance in radar multiple echo classification is solved, improving classification accuracy and computational efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-29
- Publication Date
- 2026-04-03
AI Technical Summary
In existing radar multiple echo classification methods, the uneven number of samples for each echo category leads to the inability to achieve stable classification results for minority classes, resulting in an unsatisfactory overall classification accuracy.
The EasyEnsemble algorithm is adopted to merge all multiple echo categories into one class. Combined with the undersampling method, a radar multiple echo binary classifier is designed. The AdaBoost classifier is trained using logarithmic amplitude and phase data for classification.
It improves the accuracy of radar multi-echo classification, reduces computational load, and enhances radar data processing and detection performance.
Smart Images
Figure CN116047444B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar signal processing technology, specifically to a radar multiple echo binary classification method based on the EasyEnsemble algorithm. Background Technology
[0002] Radar detects and ranges targets by emitting electromagnetic waves to illuminate them and receiving their echoes. Pulse radar periodically emits signals and receives echoes, transmitting multiple pulses into space with a certain pulse repetition period. During each pulse reception period, it receives the echo signal reflected by a target in space. When the target is far from the radar, exceeding its maximum detection range (maximum unambiguous range), the echo signal reflected by the target is not received during the current pulse reception period but is received during a subsequent pulse reception period. If the target's distance from the radar does not exceed the radar's maximum unambiguous range, the echo signal reflected by the target can be received during the current pulse reception period. When the radar emits the first pulse, if the echo signal reflected by a target in space is received during the first pulse reception period, it is called a first echo; if the echo signal is received during the second pulse reception period, it is called a second echo; and if the echo signal is received during the third pulse reception period, it is called a third echo. Similarly, if the echo signal reflected by a target in space is received during the Nth pulse reception signal, it is called the Nth echo. Multiple echoes received by the radar will be detected as target traces in subsequent target detection processes. However, these traces are not of interest to the current detection radar. Excessive redundancy of traces increases the computational load of radar data processing, hindering subsequent operations.
[0003] Most existing methods classify radar multiple echoes by comparing an empirical threshold with the logarithmic amplitude difference between adjacent pulses. When the logarithmic amplitude difference between adjacent pulses exceeds the empirical threshold, it is classified as a multiple echo. However, in practical applications, target categories are complex and diverse, and the amplitude fluctuations between adjacent pulses are significant. The threshold value for multiple echo classification is difficult to determine, resulting in low accuracy when using empirical thresholds to classify multiple echoes, and failing to significantly alleviate the problem of excessive computation. Furthermore, most existing classification methods are designed for multi-class problems, classifying multiple echoes into different categories based on their number of echoes. In real-world scenarios, the number of multiple echoes of different categories in detected points often varies considerably, facing an imbalance in the number of samples for single, double, and triple echoes. This causes the designed multi-class algorithms to favor echo categories with more training samples, performing poorly on categories with very few training samples. This results in echo categories with very few samples being overwhelmed by those with a large number of samples, affecting the overall classification accuracy and failing to achieve good classification results. This results in a large number of poor-quality dots remaining in the detected dots, which is detrimental to improving the performance of radar data processing. Summary of the Invention
[0004] Technical problems to be solved
[0005] To address the problem of unstable classification performance for minority classes due to the imbalanced number of samples in each echo category in existing multi-class classification models, resulting in unsatisfactory overall classification accuracy, this invention provides a radar multi-echo binary classification method. This method merges all multiple echo categories during data preprocessing, distinguishes them from single echoes, and incorporates the EasyEnsemble algorithm. This solves the problem of the imbalanced number of echo samples affecting the final classification performance, thereby improving the classification accuracy of radar multi-echo systems.
[0006] Technical solution
[0007] A radar multiple echo binary classification method based on the EasyEnsemble algorithm, characterized by the following steps:
[0008] S1: First, the radar multipulse signal echo is received by the radar receiver. After preprocessing operations such as channel compensation, beamforming, and pulse compression, the complex data X of the radar multipulse echo is obtained. N×L = [x1, x2, ... x l , ...x L ] N×L Where N represents the number of pulses received by the radar, and L represents the number of range cells included in the radar data. This represents the column vector consisting of N pulse echoes from the l-th range cell in the radar echo.
[0009] S2: Analyzing radar multipulse echo complex data X using radar target detection methods. N×L = [x1, x2, ..., x L ] N×L The data is processed to obtain radar detection point data y. M×1 = [y1, y2, ... y m , ...y M ] M×1 This includes traces of targets reflected from a single echo and traces of targets reflected from multiple echoes; where M represents the number of radar-detected traces, and y m m = 1, 2, ..., M represents the range cell number of the m-th radar point;
[0010] S3: Based on the radar detection point data y in S2 M×1 = [y1, y2, ..., y M ] M×1 The range cell is located in, from the radar multipulse echo complex data X in S1. N×L = [x1, x2, ..., x L ] N×L Extract the radar echo complex data of N pulses at the corresponding range cell and convert it into logarithmic amplitude data S′. N×M =[s′1,s′2,…,s′ N ] N×M With phase data S″ N×M =[s″1,s″2,…,s″ N ] N×M , merge S′ N×M With S″ N×M After transposing, store as S M×2N =[S′ N×M ,S″ N×M ] T The radar dot category k is determined by empirical threshold judgment and manual assistance. M×1 = [k1, k2, ... k m , ...k M ] M×1 , where k m m = 1, 2, ..., M is a value between 1 and N, representing the current mark category, also known as the sample label;
[0011] S4: Logarithmic amplitude and phase data of the N pulses of the radar detection point obtained in S3. M×2N Preprocessing is performed based on the point category to which it belongs, merging all multiple echo categories into one category, and setting the label of the merged category to 0. The resulting preprocessed point category label is k′. M×1 = [k′1, k′2, ... k′] m,…k′ M ] M×1 , where k′ m m = 1, 2, ..., M is 0 or 1, and the number of samples in the two classes is still unbalanced;
[0012] S5: Design a radar multiple echo binary classifier based on the EasyEnsemble algorithm. The input is the logarithmic amplitude data and phase data S of N pulses from the radar detection point. M×2N The output is the category k to which the radar detected point belongs. M×1 = [k1, k2, ..., k M ] M×1 The EasyEnsemble algorithm designed in this invention contains 10 AdaBoost classifiers, each of which consists of 50 decision tree base classifiers; the results of all AdaBoost classifiers are weighted and summed, and the final output class is determined by the sign function;
[0013] S6: Utilize the logarithmic amplitude data and phase data S of the N pulses corresponding to the range cell of the radar spot in S3. M×2N and sample label k′ M×1 =[k′1,k′2,…,k′ M ] M×1 Train the classifier designed in S5 and verify its performance using a test set.
[0014] A further technical solution of the present invention: The sampling method of the EasyEnsemble algorithm described in S5 is Bootstrap sampling. The basic idea is to perform sampling with replacement on the majority class sample set N to generate several sample subsets N. i Where i = 1, 2, ..., T is the number of base classifiers, and N i The number of samples in N is the same as the number of samples in the minority class sample set R; i Combined with R, it is used as training data input for each base classifier to learn.
[0015] A further technical solution of the present invention: S6 is as follows:
[0016] S61: The logarithmic amplitude and phase data S of the N pulses corresponding to the preprocessed radar spot data in S4 at the range cell. M×2N The training dataset is randomly divided along the distance dimension according to the p:q ratio. and test dataset Two parts; where p + q = 1, p > q. Q = MP, where P is the number of training data points and Q is the number of test data points. Indicates rounding down;
[0017] S62: Based on the logarithmic amplitude and phase data of N pulses at the range cell corresponding to the preprocessed radar detection point at S4, the training dataset is obtained by using empirical thresholds and manual assistance to determine the sample label corresponding to the detection point. Corresponding sample label vector and test dataset Corresponding sample label vector in This represents the sample category label of the detection point corresponding to the p-th sample in the training dataset. If the point belongs to the target point of a single echo reflection, the sample label is 1; if the point belongs to the target point of multiple echo reflections, the sample label is 0. This represents the sample category label of the detection point corresponding to the q-th sample in the test dataset;
[0018] S63: Training dataset partitioned using S61 The sample class labels corresponding to the training dataset obtained by S62 training dataset As the input to the radar multiple echo binary classifier based on the EasyEnsemble algorithm designed in S5, the output is the corresponding sample category label;
[0019] S64: Test dataset partitioned using S61 The sample category labels corresponding to the test dataset obtained from S0002 The classifier based on the EasyEnsemble algorithm obtained by S63 is tested, the accuracy and loss rate of the classifier on the test set are calculated, and the network model is optimized based on the test results.
[0020] S65: The optimized classifier based on the EasyEnsemble algorithm is used for binary classification of radar multiple echoes.
[0021] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.
[0022] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.
[0023] Beneficial effects
[0024] The present invention provides a radar multiple echo binary classification method based on the EasyEnsemble algorithm, which has the following advantages:
[0025] 1. In the data preprocessing stage, this invention merges all multiple echoes into one class, transforming it into a binary classification problem to distinguish between single radar echoes and multiple echoes. In practical engineering applications, only target points with single echoes are of interest to the current detection radar; therefore, this operation can effectively improve classification accuracy and computational efficiency.
[0026] 2. This invention introduces phase information into the existing method that only uses amplitude information after pulse compression to classify radar multiple echoes. This allows the designed classification algorithm to learn more feature information from the input data, thereby further improving the classification accuracy of radar multiple echoes.
[0027] 3. This invention is a radar multiple echo binary classification method based on the EasyEnsemble algorithm. It is the first to combine undersampling methods from machine learning with ensemble learning classification algorithms for radar multiple echo binary classification. The majority class sample set after merging all multiple echo categories is sampled with replacement to obtain multiple balanced positive and negative sample subsets to train multiple AdaBoost classifiers. The classification results are then weighted and summed, and the sign function is used to determine the radar multiple echo category. Compared to existing radar multiple echo classification methods based on empirical thresholds, this invention further improves classification accuracy and does not rely on empirical thresholds. This invention represents an important processing step in the field of radar data processing, improving detection performance while reducing the computational load of radar target detection algorithms.
[0028] 4. The EasyEnsemble algorithm used in this invention is a hybrid ensemble strategy that incorporates both Bagging and Boosting ensemble ideas, resulting in a model with excellent robustness and strong generalization ability. The EasyEnsemble algorithm overcomes the drawback of undersampling, which may lead to the loss of important classification information, effectively improving the classification accuracy of minority class samples. Attached Figure Description
[0029] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.
[0030] Figure 1 This is a schematic diagram of the implementation process of the present invention;
[0031] Figure 2 This is a schematic diagram of the algorithm flow of the AdaBoost classifier used in this invention. Detailed Implementation
[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0033] See Figure 1 This invention is a radar multiple echo binary classification method based on the EasyEnsemble algorithm, comprising the following methods:
[0034] S1: First, the radar multipulse signal echo is received by the radar receiver. After preprocessing operations such as channel compensation, beamforming, and pulse compression, the complex data X of the radar multipulse echo is obtained. N×L = [x1, x2, ..., x L ] N×L Where N represents the number of pulses received by the radar, and L represents the number of range cells included in the radar data. This represents the column vector consisting of N pulse echoes from the l-th range cell in the radar echo.
[0035] S2: Analyzing radar multipulse echo complex data X using radar target detection methods. N×L = [x1, x2, ..., x L ] N×L The data is processed to obtain radar detection point data y. M×1 = [y1, y2, ..., y M ] M×1 This includes traces of targets reflected from a single echo and traces of targets reflected from multiple echoes; where M represents the number of radar-detected traces, and y m m = 1, 2, ..., M represents the range cell number of the m-th radar point;
[0036] S3: Based on the radar detection point data y in S2 M×1 = [y1, y2, ..., y M ] M×1 The range cell is located in, from the radar multipulse echo complex data X in S1. N×L = [x1, x2, ..., x L ] N×L Extract the radar echo complex data of N pulses at the corresponding range cell and convert it into logarithmic amplitude data S′. N×M =[s′1,s′2,…,s′ N ] N×M With phase data S″ N×M =[s″1,s″2,…,s″ N] N×M , merge S′ N×M With S″ N×M After transposing, store as S M×2N =[S′ N×M ,S″ N×M ] T The radar dot category k is determined by a combination of empirical threshold judgment and manual assistance. M×1 = [k1, k2, ..., k M ] M×1 , where k m m = 1, 2, ..., M is a value between 1 and N, representing the current mark category, also known as the sample label;
[0037] S4: Logarithmic amplitude and phase data of the N pulses of the radar detection point obtained in S3. M×2N Preprocessing is performed based on the point category to which it belongs, merging all multiple echo categories into one category, and setting the label of the merged category to 0. The resulting preprocessed point category label is k′. M×1 =[k′1,k′2,…,k′ M ] M×1 , where k′ m m = 1, 2, ..., M is 0 or 1, and the number of samples in the two classes is still unbalanced.
[0038] S5: Design the EasyEnsemble algorithm, with the input being the logarithmic amplitude data and phase data S of N pulses from the radar detection point. M×2N The output is the category k to which the radar detected point belongs. M×1 = [k1, k2, ..., k M ] M×1 ;
[0039] S6: Utilize the logarithmic amplitude data and phase data S of the N pulses corresponding to the range cell of the radar spot in S3. M×2N and sample label k′ M×1 =[k′1,k′2,…,k′ M ] M×1 Train the classifier designed in S5 and verify its performance using a test set;
[0040] S7: Based on the EasyEnsemble algorithm obtained in S5, a radar multiple echo binary classification method is proposed. This classification method is applied to the radar signal processing process and can effectively suppress multiple echoes received by the radar, improve the quality of radar detection points, and thus enhance the tracking performance of radar data processing.
[0041] The data preprocessing described in S4 also includes the following methods:
[0042] S01: Due to the significant differences in the number of multiple echoes in the original sample set, the designed classification algorithm often struggles to effectively distinguish between the minority classes, which constitute a very small percentage of the samples. However, in practical engineering applications, only the target point trace from a single echo is of interest to the current detection radar. Therefore, this invention merges all multiple echoes into one class, and then distinguishes between single and multiple echoes based on this class. After this operation, the radar multiple echo classification algorithm is transformed into a binary classification problem.
[0043] The EasyEnsemble algorithm designed in S5 also includes the following methods:
[0044] S001: The sampling method of the designed EasyEnsemble algorithm is Bootstrap sampling. The basic idea is to sample the majority class sample set N with replacement to generate several sample subsets N. i Where i = 1, 2, ..., T is the number of base classifiers, and N i The number of samples in N is the same as the number of samples in the minority class sample set R. i Combined with R, it is used as training data input for each base classifier to learn.
[0045] S002: The EasyEnsemble algorithm uses AdaBoost as its base classifier. AdaBoost is an ensemble learning algorithm employing a boosting strategy. The training process is hierarchical, with the base models trained sequentially. If a data point is misclassified in one iteration, it will be given a greater weight in the next training iteration. AdaBoost's base classifier is typically a decision tree with a maximum depth of 1 by default.
[0046] S003: The EasyEnsemble algorithm designed uses a weighted summation of the results from all AdaBoost classifiers, and then uses the sign function to determine the class. This ensemble of multiple classifiers reduces variance and improves the model's robustness.
[0047] Using the logarithmic amplitude data and phase data S of N pulses corresponding to the range cell of the radar spot in S3 M×2N and sample label k′ M×1 =[k′1,k′2,…,k′ M ] M×1 The EasyEnsembl classifier designed in S5 is trained, and its performance is verified using a test set; the following methods are also included:
[0048] S0001: The logarithmic amplitude and phase data of the N pulses corresponding to the preprocessed radar spot data in S4 at the range cell are...M×2N The training dataset is randomly divided along the distance dimension according to the p:q ratio. and test dataset Two parts; where p + q = 1, p > q. Q = MP, where P is the number of training data points and Q is the number of test data points. Indicates rounding down;
[0049] S0002: Based on the logarithmic amplitude and phase data of N pulses at the range cell corresponding to the preprocessed radar detection point at S4, the training dataset is obtained by using empirical thresholds and manual assistance to determine the sample label corresponding to the detection point. Corresponding sample label vector and test dataset Corresponding sample label vector in This represents the sample category label of the detection point corresponding to the p-th sample in the training dataset. If the point belongs to the target point of a single echo reflection, the sample label is 1; if the point belongs to the target point of multiple echo reflections, the sample label is 0. This represents the sample category label of the detection point corresponding to the q-th sample in the test dataset;
[0050] S0003: Training dataset partitioned using S0001 The sample class labels corresponding to the training dataset obtained from S0002 training dataset As the input to the radar multiple echo binary classifier based on the EasyEnsemble algorithm designed in S5, the output is the corresponding sample category label. The EasyEnsemble algorithm designed in this invention contains 10 AdaBoost classifiers, each of which consists of 50 decision tree-based classifiers. The results of all AdaBoost classifiers are weighted and summed, and the final output category is determined by the sign function.
[0051] S0004: Test dataset partitioned using S0001 The sample category labels corresponding to the test dataset obtained from S0002 The classifier based on the EasyEnsemble algorithm obtained in S0003 was tested, and the accuracy and loss rate of the classifier on the test set were calculated. The network model was then optimized based on the test results.
[0052] S0005: The optimized classifier based on the EasyEnsemble algorithm is used for binary classification of radar multiple echoes.
[0053] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.
Claims
1. A radar multiple echo binary classification method based on the EasyEnsemble algorithm, characterized in that... The steps are as follows: S1: First, the radar multipulse signal echo is received by the radar receiver. After preprocessing operations such as channel compensation, beamforming, and pulse compression, the radar multipulse signal echo complex data is obtained. ;in This indicates the number of pulses received by the radar. This indicates the number of range cells included in the radar data. Indicates the first in the radar echo distance units A column vector consisting of pulse echoes; S2: Using radar target detection methods to process radar multipulse echo data. The data is processed to obtain radar detection point data. This includes the traces of a target reflected from a single echo and the traces of a target reflected from multiple echoes; among which This indicates the number of radar detection points. Indicates the first The range cell number of each radar spot; S3: Based on the radar detection point data in S2 The range cell is located in, from the radar multipulse echo complex data in S1. Extract the corresponding distance cell The radar echo complex data of each pulse is converted into logarithmic amplitude data. With phase data ,merge and Post-transposed storage as The radar dot category is determined by a combination of empirical threshold judgment and manual assistance. ,in for The values between these ranges represent the current category of the trace, also known as the sample label; S4: Radar detection points obtained in S3 Logarithmic amplitude and phase data of each pulse Preprocessing is performed based on the point category to which it belongs. All multiple echo categories are merged into one category, and the label of the merged category is set to 0. The resulting preprocessed point category label is: ,in for or At this point, the number of samples in the two classes is still imbalanced; S5: Design a radar multiple echo binary classifier based on the EasyEnsemble algorithm, with the input being radar detection points. Logarithmic amplitude data and phase data of each pulse The output is the category to which the radar-detected point belongs. The EasyEnsemble algorithm contains 10 AdaBoost classifiers, each consisting of 50 decision tree-based classifiers; the results of all AdaBoost classifiers are weighted and summed. The function determines the final output category; S6: Utilize the radar spot corresponding to the range cell in S3 Logarithmic amplitude data and phase data of each pulse and sample labels Train the classifier designed in S5 and verify its performance using a test set.
2. The radar multiple echo binary classification method based on the EasyEnsemble algorithm according to claim 1, characterized in that... The sampling method of the EasyEnsemble algorithm described in S5 is Bootstrap sampling, the basic idea of which is to sample the majority class sample set. Perform sampling with replacement to generate several sample subsets. ,in The number of base classifiers, The number of samples and the minority class sample set The number of samples in each sample is the same; and These data are combined and used as training data input to each base classifier for learning.
3. The radar multiple echo binary classification method based on the EasyEnsemble algorithm according to claim 1, characterized in that... S6 is detailed below: S61: The preprocessed radar spot data from S4 is mapped to the corresponding range cell. Logarithmic amplitude and phase data of each pulse Along the distance dimension The proportion of the dataset is randomly divided into training datasets. and test dataset Two parts; of which , , , It is the number of training datasets. It is the number of test datasets. Indicates rounding down; S62: Based on the range cell corresponding to the preprocessed radar detection point at S4. The training dataset is obtained by using the logarithmic amplitude and phase data of each pulse, along with empirical thresholds and manual assistance to determine the sample labels corresponding to the detected points. Corresponding sample label vector and test dataset Corresponding sample label vector ;in Indicates the first in the training dataset The sample category label for each detection point is determined by the number of samples. If the point belongs to a target point of a single echo reflection, then the sample label is [label missing]. If the mark belongs to a target mark that has undergone multiple echo reflections, then the sample label is... ; Indicates the first in the test dataset The sample category label for each sample corresponding to the detection point; S63: Training dataset partitioned using S61 The sample class labels corresponding to the training dataset obtained from S62 , training dataset As the input to the radar multiple echo binary classifier based on the EasyEnsemble algorithm designed in S5, the output is the corresponding sample category label; S64: Test dataset partitioned using S61 The sample class labels corresponding to the test dataset obtained in S2 The classifier based on the EasyEnsemble algorithm obtained by S63 is tested, the accuracy and loss rate of the classifier on the test set are calculated, and the network model is optimized based on the test results. S65: The optimized classifier based on the EasyEnsemble algorithm is used for binary classification of radar multiple echoes.
4. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of claim 1.
5. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method of claim 1.