Discrete electromagnetic signal sample playback method and device based on clustering and density evaluation

Abstract

The present invention relates to the fields of digital signal processing and deep learning, and provides a method and device for replaying discrete electromagnetic signal samples based on clustering and density assessment. The method comprises: preprocessing discrete electromagnetic signal samples; clustering and importance assessment of the preprocessed samples to obtain an ordered replay sample set; density assessment of the samples in the ordered replay sample set to calculate a comprehensive importance score; screening the ordered replay sample set based on the comprehensive importance score; and resampling the screening results under density constraints to generate a replay sample set. The present invention can improve the representativeness of each group of samples, thereby alleviating problems such as decreased model generalization ability and forgetting of old classes in incremental learning scenarios, and improving the stability and accuracy of incremental learning.

Description

Technical Field

[0001] The present invention relates to the fields of digital signal processing and deep learning, and in particular to a method and device for replaying discrete electromagnetic signal samples based on clustering and density evaluation. Background Art

[0002] Among existing Class Incremental Learning (CIL) methods, the most common replay strategies include random sampling and representative sampling based on category centers. While simple and easy to implement, random sampling (such as reservoir sampling) often stores a large number of redundant and information-deficient samples, resulting in a large amount of duplicate data in the memory bank and a failure to fully cover marginal and weak areas of the data distribution. This affects the model's adaptability to extreme samples and exacerbates the problem of catastrophic forgetting.

[0003] Sampling methods based on category centers (such as center prototype playback) reduce storage overhead by storing the centroid or prototype of each category. However, these methods tend to ignore samples near category boundaries and fail to effectively handle inter-category interference, leading to more misclassifications and forgetting when prototypes of different categories are close together. Research has shown that finding the center of high-density regions through techniques such as Attentional Mean-Shift can improve prototype representativeness, but there are still significant shortcomings in coverage of edge samples and generation of synthetic data.

[0004] To improve the diversity and representativeness of replay sets, some work has introduced density clustering methods (such as density-space-based clustering algorithms) to identify core samples from high-density clusters and assess sample importance using histograms or kernel density estimation. However, these methods remain largely theoretical or experimental, lacking a comprehensive, systematic process and deployable device implementation. This makes it difficult to achieve sample reduction in dense areas and oversampling of edge samples while maintaining a controlled number of replay samples.

[0005] In addition, previous research has explored sample selection schemes that combine multi-strategy fusion with active learning to achieve a balance between new and old categories. However, multi-strategy models often have high computational complexity, making them difficult to meet the needs of real-time or resource-constrained scenarios such as electromagnetic signal classification. Experiments have shown that randomly sampled replay samples, even when derived from an accurate diffusion model, can significantly degrade performance. Selection methods based on mutual information or uncertainty are sensitive to storage and computational costs and cannot be implemented efficiently with limited resources.

[0006] In application scenarios with discrete distributions such as complex electromagnetic signals, existing playback methods usually only retain category center samples, ignoring the dynamic balance between densely distributed areas and sparse edge areas in the signal feature space. They are unable to simultaneously identify and preserve common and rare patterns, resulting in insufficient recognition of edge patterns and poor adaptability to new categories when the model learns or retrains new categories.

[0007] In summary, existing incremental learning replay strategies are difficult to achieve unified sample extraction and replay schemes between dense areas and edge areas while keeping storage and computational costs limited. They lack a systematic, adjustable, and complete process implementation solution that takes into account multiple distribution characteristics. Summary of the Invention

[0008] The present invention aims to provide a discrete electromagnetic signal sample playback method and device based on clustering and density assessment to solve the problem of unreasonable playback sample selection strategy in existing category incremental learning scenarios. In particular, traditional random sampling and representative sampling strategies based on category centers cannot take into account both the representativeness and edge diversity of the data when processing samples with highly discrete characteristics or uneven distribution, which in turn leads to catastrophic forgetting of the model, decreased adaptability to new and old categories, and unstable generalization performance.

[0009] In a first aspect, the present invention provides a discrete electromagnetic signal sample playback method based on clustering and density evaluation, comprising:

[0010] Preprocessing samples of discrete electromagnetic signals;

[0011] Perform clustering and importance evaluation on the preprocessed samples to obtain an ordered playback sample set;

[0012] Perform density evaluation on samples in the ordered playback sample set and calculate the comprehensive importance score;

[0013] Screen the ordered playback sample set based on the comprehensive importance score;

[0014] The screening results are resampled under density constraints to generate a playback sample set.

[0015] In some embodiments, preprocessing the samples of the discrete electromagnetic signal refers to normalizing the feature vectors of each category in the samples of the original discrete electromagnetic signal.

[0016] In some embodiments, the normalization process includes:

[0017] A sample of a discrete electromagnetic signal represents a set of pulse-to-pulse parameters, including pulse repetition interval, carrier frequency, and pulse width;

[0018] The pulse repetition interval, carrier frequency and pulse width are normalized using their respective minimum and maximum values.

[0019] In some embodiments, the minimum and maximum values ​​are determined based on the data distribution of the data set, specifically including:

[0020] Get the minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width in the data set;

[0021] Expand the minimum and maximum values ​​of the pulse repetition interval, carrier frequency, or pulse width in the data set according to actual application requirements;

[0022] The minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width in the sample frequency band and the expanded data set are used to calculate the corresponding minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width.

[0023] In some embodiments, clustering and importance assessment of the preprocessed samples includes:

[0024] Use K-Means strategy to initialize cluster centers to ensure that cluster centers are evenly distributed;

[0025] Iteratively perform sample attribution updates and cluster center recalculations until the cluster center position changes within the required range or the maximum number of iterations is reached, and obtain the cluster label of each sample and the number of samples in each cluster;

[0026] By using histogram and joint density distribution methods to evaluate the importance of sample types, clustered compact samples and edge samples are selected from the cluster labels of samples. The clustered compact samples are placed at the front and the edge samples are placed at the back to obtain an ordered playback sample set.

[0027] In some embodiments, performing density assessment on samples in the ordered playback sample set and calculating a comprehensive importance score includes:

[0028] In each cluster, the Gaussian kernel function is used to estimate the sample density, and the sample density value of the kernel density function of the clustered compact samples and edge samples in the ordered playback sample set is obtained;

[0029] Combined with the sample density value and the distance from the sample to the cluster center, the comprehensive importance score is calculated according to the preset weight.

[0030] In some embodiments, screening the ordered playback sample set according to the comprehensive importance score includes:

[0031] Arrange the samples in the ordered playback sample set in descending order according to the comprehensive importance score, and extract the first proportion of samples after descending order as the high-importance sample set;

[0032] After filtering out the high-importance sample set, the remaining samples are sorted in ascending order of comprehensive importance scores and the second-highest proportion of samples are extracted as the edge sample set;

[0033] The high-importance sample set and the marginal sample set are merged in sequence to form a candidate replay list.

[0034] In some embodiments, resampling the screening results under density constraints comprises:

[0035] Resample the candidate replay list: undersample the high-importance sample set and oversample the marginal sample set;

[0036] Merge the resampling results of all categories to obtain the playback dataset and ensure that the total number of samples in each category meets the overall capacity constraint;

[0037] The replay dataset is sorted again by the comprehensive importance score and the final replay sample set is output.

[0038] In some embodiments, the ratio of undersampling to oversampling can be dynamically adjusted based on system storage and computing resources.

[0039] In a second aspect, the present invention provides a discrete electromagnetic signal sample playback device based on clustering and density evaluation, comprising:

[0040] A first processing unit, configured to pre-process samples of discrete electromagnetic signals;

[0041] The second processing unit is used to cluster and evaluate the importance of the preprocessed samples to obtain an ordered playback sample set;

[0042] The third processing unit is used to perform density evaluation on samples in the ordered playback sample set and calculate a comprehensive importance score;

[0043] a fourth processing unit, configured to screen the ordered playback sample set according to the comprehensive importance score;

[0044] The fifth processing unit is used to resample the screening results under density constraints to generate a playback sample set.

[0045] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0046] The replay of discrete electromagnetic signal samples directly affects the training effect in the incremental learning scenario. Different from the existing sample replay ideas, the present invention effectively combines the characteristic distribution trend of the used data, combines the actual distribution state of the electromagnetic signal in the actual engineering scenario, constructs the importance sampling strategy for the previous and next data, and improves the sampling weight of the edge data. In terms of data preprocessing, taking into account the data characteristics of the uneven distribution of the signal itself, the improved K-Means clustering method is used to improve the representativeness of each group of samples, thereby alleviating the problems of decreased model generalization ability and forgetting of old classes in the incremental learning scenario, and improving the stability and accuracy of incremental learning. Therefore, the advantages of the present invention compared with the prior art are:

[0047] 1. Compared with random sampling and representative sampling based on category centers, this method adopts the sampling idea of ​​data distribution, focusing more on the impact of edge samples on incremental learning. This avoids the problem of marginal sample loss after a large number of samples are introduced in incremental learning, which leads to a decrease in generalization ability.

[0048] 2. Compared with the sampling method using density clustering, an importance scoring strategy is introduced to sort different clustering results within the class and sample according to the degree of importance, thus avoiding the problems of small inter-class differences after ordinary sampling, lack of representativeness of samples after clustering, and weak distinguishability.

[0049] By combining the actual open set electromagnetic scene distribution characteristics in actual engineering scenarios, the present invention can improve the representativeness of each group of samples, thereby alleviating problems such as the decline in model generalization ability and forgetting of old classes in incremental learning scenarios, and improving the stability and accuracy of incremental learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a discrete electromagnetic signal sample playback method based on clustering and density evaluation is provided in an embodiment of the present invention.

[0051] Figure 2a This is a complete old class data distribution diagram before sampling in an embodiment of the present invention.

[0052] Figure 2b This is a complete old class data distribution diagram after sampling in an embodiment of the present invention.

[0053] Figure 2c This is a complete new class data distribution diagram before sampling in an embodiment of the present invention.

[0054] Figure 2d This is a complete new class data distribution diagram after sampling in an embodiment of the present invention.

[0055] Figure 3a This is a data distribution diagram of the old category 5 before sampling in an embodiment of the present invention.

[0056] Figure 3bThis is a data distribution diagram of the fifth category of the old category after sampling in an embodiment of the present invention.

[0057] Figure 3c This is a data distribution diagram of the 12th category of the old category before sampling in an embodiment of the present invention.

[0058] Figure 3d This is the data distribution diagram of the 12th category of the old category after sampling in an embodiment of the present invention.

[0059] Figure 3e This is a data distribution diagram of the old category 4 before sampling in an embodiment of the present invention.

[0060] Figure 3f This is a data distribution diagram of the old category 4 after sampling in an embodiment of the present invention.

[0061] Figure 4a Graph showing the forgetting curves for different numbers of old and new classes in an embodiment of the present invention.

[0062] Figure 4b This is an accuracy curve diagram for different numbers of old classes and new classes in an embodiment of the present invention.

[0063] Figure 5 A schematic structural diagram of a discrete electromagnetic signal sample playback device based on clustering and density evaluation provided by an embodiment of the present invention.

[0064] Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention.

[0065] In the accompanying figure: Class represents category, Old represents old class, and New represents new class. DETAILED DESCRIPTION

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0067] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0068] Example

[0069] like Figure 1As shown, an embodiment of the present invention provides a discrete electromagnetic signal sample playback method based on clustering and density evaluation, comprising:

[0070] S1, preprocessing the samples of discrete electromagnetic signals;

[0071] S2, clustering and importance evaluation of the preprocessed samples to obtain an ordered playback sample set;

[0072] S3, perform density evaluation on samples in the ordered playback sample set and calculate the comprehensive importance score;

[0073] S4, screening the ordered playback sample set according to the comprehensive importance score;

[0074] S5, resample the screening results under density constraints to generate a playback sample set.

[0075] Therefore, the present invention integrates cluster distribution structure and sample density information, and combines the sample selection method with the density adjustment extraction mechanism to achieve adjustable, efficient, and edge-sensitive discrete electromagnetic signal sample playback, significantly improving the fusion ability between new and old knowledge in the incremental learning process, learning stability, and the comprehensive performance of the final model.

[0076] The following is a detailed description of a discrete electromagnetic signal sample playback method based on clustering and density evaluation provided by an embodiment of the present invention. The specific steps are as follows:

[0077] S1, preprocessing the samples of discrete electromagnetic signals.

[0078] In some embodiments, preprocessing the discrete electromagnetic signal samples refers to normalizing the feature vectors of each category in the original discrete electromagnetic signal samples to eliminate scale differences between different dimensions.

[0079] The normalization process specifically includes:

[0080] Input discrete electromagnetic signal sample sequence ,in, n is the number of samples in the sample sequence, the sample sequence The i samples represents a set of inter-pulse parameters, Indicates the i The pulse repetition interval of samples, Indicates the i The carrier frequency of the samples, Indicates the i The pulse width of each sample.

[0081] For each sample Use the corresponding minimum and maximum values ​​to perform normalization to obtain the preprocessed samples The specific normalization formula is as follows:

[0082]

[0083]

[0084]

[0085] Among them, the minimum carrier frequency , maximum carrier frequency , minimum pulse repetition interval , maximum pulse repetition interval , minimum pulse width , maximum pulse width The value of comes from the data distribution of the dataset.

[0086] Carrier frequency For example, first get the minimum value of the carrier frequency in the data set and maximum value , and adjust the minimum value of the carrier frequency according to the actual application requirements and maximum value The difference is expanded by a certain proportion. In this embodiment It is 1 / 4 of the ratio of the current sample value range to the sample frequency band range. The specific formula is as follows:

[0087]

[0088]

[0089]

[0090] in, Indicates the maximum value of the sample frequency band. Indicates the minimum value of the sample frequency band.

[0091] Similarly, the minimum pulse repetition interval , maximum pulse repetition interval , minimum pulse width , maximum pulse width The value is also determined based on the data distribution of the data set. The minimum carrier frequency obtained during the sample training phase , maximum carrier frequency , minimum pulse repetition interval , maximum pulse repetition interval , minimum pulse width , maximum pulse width The data preprocessing that will be synchronized to the real-time recognition processing will be normalized with the same maximum and minimum values ​​to ensure the consistency of features after parameter normalization in the training and recognition stages.

[0092] S2, clustering and importance evaluation are performed on the preprocessed samples to obtain an ordered playback sample set.

[0093] In some embodiments, the clustering is performed using an improved K-Means clustering algorithm within each category, and step S2 specifically includes:

[0094] Use K-Means strategy to initialize cluster centers to ensure that cluster centers are evenly distributed;

[0095] Iteratively perform sample attribution updates and cluster center recalculations until the cluster center position changes within the required range (usually the cluster center position needs to change slightly) or the maximum number of iterations is reached, and the cluster labels of all samples are obtained. And the sample number statistics of each cluster.

[0096] The importance of sample types is evaluated by using histogram and joint density distribution methods, and clustered and compact samples are selected from the cluster labels of the samples. and marginal samples (samples at the edge of the distribution) ; Place the clustered compact samples at the front end and the edge samples at the back end to obtain an ordered playback sample set , expressed as:

[0097]

[0098] in, Represents a sorting function.

[0099] S3, performs density evaluation on samples in the ordered playback sample set and calculates the comprehensive importance score.

[0100] In each cluster, the Gaussian kernel function is used to estimate the sample density and obtain an ordered playback sample set. Inner clustered compact samples and edge samples The sample density value of the kernel density function , expressed as:

[0101]

[0102] in, Represents clustered compact samples and edge samples The number of samples, Indicates the i The cluster labels of samples, Represents the cluster labels of all samples Except the i Cluster markers outside of samples, represents the smoothing bandwidth, Represents the kernel function.

[0103] Combined sample density values And the distance from the sample to the cluster center , calculate the comprehensive importance score according to the preset weight α and (1-α) , expressed as:

[0104]

[0105] in, Represents the normalization function.

[0106] S4, screen the ordered playback sample set according to the comprehensive importance score.

[0107] The samples in the ordered playback sample set are divided into Arrange in descending order, and extract the first proportion after descending order as the high-importance sample set ;

[0108] After filtering out the high-importance sample set, the remaining samples are sorted according to the comprehensive importance score. Extract the second largest proportion of samples in ascending order as the edge sample set ;

[0109] The high importance sample sets are sequentially With marginal sample set Merge to form categories Candidate replay list .

[0110] S5, resample the screening results under density constraints to generate a playback sample set.

[0111] Resample the candidate replay list: resample the high-importance sample set Undersample (extraction ratio β) and Oversampling is performed (synthesis ratio γ); wherein the ratio of undersampling to oversampling can be dynamically adjusted based on system storage and computing resources to balance the number of samples and coverage quality.

[0112] Combine the resampling results of all categories to obtain the playback dataset, and ensure that the total number of samples in each category meets the overall capacity constraint;

[0113] Replay dataset Sort again by comprehensive importance score and output the final playback sample set , used for subsequent incremental training or model fine-tuning stages.

[0114] Through the above-mentioned step-by-step implementation, the present invention can retain important representative samples in densely distributed areas while taking into account rare samples at the edge of the distribution under limited storage and computing conditions, thereby achieving a balance between sample diversity and edge coverage and improving the stability and accuracy of incremental learning.

[0115] To verify the effectiveness of the present invention, the present invention conducts comparative evaluation on incremental training examples of radar target classification models. The experimental conditions are as follows:

[0116] Classification dataset: A simulated small-sample radar dataset is used, which contains 12 old classes, with 2,000 signals of length 150,000 in each class. After segmentation, the old class samples are approximately 24,000. The newly added new class signals are also segmented and processed.

[0117] Classification model training configuration: The network uses the Adam optimizer, with an initial learning rate of 0.0001, and the accuracy of the round and validation set decreases by 50% per step; after 40 rounds of pre-training, the feature extraction module is frozen, and only the classifier and playback samples are incrementally trained.

[0118] Sample playback ratio: Old class data is replayed at ratios of 2% and 10%, and new class data is replayed at ratios of 2% and 50%. During testing, the remaining 98% and 90% old class samples, as well as 98% and 50% new class samples, are used for evaluation.

[0119] (1) Data preprocessing:

[0120] Obtaining samples of the raw discrete electromagnetic signal from external storage or a data stream;

[0121] (2) Feature extraction:

[0122] All discrete electromagnetic signal samples in the current incremental batch are forward-inferred through pre-trained autoencoders or deep networks to extract intermediate coding features. As the complete set of samples to be sampled and played back, k is the number of extracted intermediate coding features, For the k intermediate encoding features.

[0123] (3) The present invention is used to sample and replay to obtain the replay sample set R required for incremental training.

[0124] (4) Incremental training and fine-tuning:

[0125] (4.1) Initial training phase: using the replay sample set The classifier is trained iteratively for several rounds on samples with medium and high importance to quickly fit the main distribution features;

[0126] (4.2) Edge fine-tuning stage: using the playback sample set Adaptively fine-tune the model based on edge samples to enhance the accuracy of rare pattern recognition and boundary discrimination;

[0127] (4.3) After each round, evaluate the performance of the validation set and dynamically adjust the extraction ratio β, synthesis ratio γ or learning rate to optimize the subsequent training effect.

[0128] (5) Model evaluation and deployment:

[0129] (5.1) Evaluate the classification accuracy, forgetting rate, and convergence speed on the full test set;

[0130] (5.2) Save the latest model parameters, update the system deployment environment, and prepare for the next incremental batch.

[0131] Figure 2a 、 Figure 2b 、 Figure 2c 、 Figure 2d This paper presents the data changes before and after the extraction of a discrete electromagnetic signal sample playback method based on clustering and density assessment. The horizontal and vertical axes are dimensionless for the convenience of visualization. The multidimensional parameters of the signal are reduced to two dimensions through principal component analysis (PCA). Figure 2b 、 Figure 2d It shows that after data extraction, good data distribution can still be guaranteed, that is, the overall data distribution status can be representatively summarized.

[0132] Figure 3a 、 Figure 3b 、 Figure 3c 、 Figure 3d 、 Figure 3e 、 Figure 3f The data changes before and after the extraction of various types of data from the data set are presented using a discrete electromagnetic signal sample playback method based on clustering and density assessment. The horizontal and vertical axes are dimensionless for the convenience of visualization. The multidimensional parameters of the signal are reduced to two dimensions through principal component analysis (PCA). Figure 3b 、 Figure 3d 、 Figure 3f This shows that each data type exhibits a state where there are fewer samples near the cluster center and more samples at the edge of the cluster. This extraction method is conducive to efficiently constructing samples that are more difficult to learn and judge, and enhances the robustness of the model.

[0133] To verify the performance of the present invention under different incremental categories and incremental samples, tests were conducted to verify the gradual increase of multiple incremental categories and incremental samples. Table 1 shows the incremental learning classification accuracy under various scenarios. To simulate a real incremental scenario, new category data was gradually added, and incremental training and comparison were performed separately. The number of newly added categories remained stable in the overall data, that is, the sample size of the new category dataset remained unchanged with each new category added.

[0134] Table 1, incremental training results:

[0135]

[0136] The accuracy curve when comparing different numbers of new classes is as follows: Figure 4a 、 Figure 4b As shown, the horizontal axis is the number of categories corresponding to different new categories in Table 1, and the vertical axis is the accuracy (unit is %). Figure 4a It shows that with the increase of new categories, although the accuracy of old category samples shows a downward trend, the accuracy still remains at a high level. From adding 1 category to adding 4 categories, the overall decrease is about 0.46%, and the category-by-category forgetting is about 0.13%. Figure 4b The results show that as the number of new categories increases, the accuracy of the new categories decreases only slightly (from adding one category to four, the overall decrease is about 2.57%, and the category-by-category decrease is about 0.67%). This shows that the present invention can improve data validity and reduce the forgetting of old categories in data-driven deep learning algorithms.

[0137] A longitudinal comparison of different old class data sets shows that when a small sample of 2% is selected from the sorted old class, the fluctuation in the number of new class samples does not drastically affect the accuracy of the old class. Specifically, after the proportion of new class samples increases from 2% to 50%, the accuracy of the old class only loses 0.15% on average, proving that when the proportion of old class playback data is low, the present invention can also ensure that the network has good classification performance.

[0138] Based on the same technical concept, such as Figure 5 As shown, the present invention provides a combined high-precision positioning method and device, comprising:

[0139] A first processing unit, configured to pre-process samples of discrete electromagnetic signals;

[0140] The second processing unit is used to cluster and evaluate the importance of the preprocessed samples to obtain an ordered playback sample set;

[0141] The third processing unit is used to perform density evaluation on samples in the ordered playback sample set and calculate a comprehensive importance score;

[0142] a fourth processing unit, configured to screen the ordered playback sample set according to the comprehensive importance score;

[0143] The fifth processing unit is used to resample the screening results under density constraints to generate a playback sample set.

[0144] The specific working principles of each processing unit in the above device can be referred to the description in the above method embodiment, which will not be repeated here.

[0145] Based on the same technical concept, an embodiment of the present invention further provides an electronic device that can implement the discrete electromagnetic signal sample playback method based on clustering and density evaluation provided in the above embodiment of the present invention. In one embodiment, the electronic device can be a server, or a terminal device or other electronic device. Figure 6 As shown, the electronic device may include:

[0146] At least one processor, and a memory connected to the at least one processor. The embodiment of the present invention does not limit the specific connection medium between the processor and the memory. Figure 6 The example in this article is that the processor and memory are connected via a bus. Figure 6 The connections between the other components are shown in bold lines, which are only for illustration and not intended to be limiting. The bus can be divided into address bus, data bus, control bus, etc. Figure 6 The processor is represented by a single thick line, but this does not mean that there is only one bus or only one type of bus. Alternatively, the processor can also be called a controller, without any limitation on the name.

[0147] In an embodiment of the present invention, the memory stores instructions executable by at least one processor. By executing the instructions stored in the memory, the at least one processor can execute the discrete electromagnetic signal sample playback method based on clustering and density evaluation discussed above.

[0148] Among them, the processor is the control center of the device, which can use various interfaces and lines to connect the various parts of the entire control device, and monitor the device as a whole by running or executing instructions stored in the memory and calling data stored in the memory, the various functions of the device and processing data.

[0149] In an optional design, the processor may include one or more processing units, and the processor may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interface, and application programs, and the modem processor primarily processes wireless communications. It is understood that the modem processor may not be integrated into the processor. In some embodiments, the processor and memory may be implemented on the same chip, or in some embodiments, they may be implemented on separate chips.

[0150] The processor can be a general-purpose processor, such as a CPU, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the methods, steps, and logic diagrams disclosed in the embodiments of the present invention. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the discrete electromagnetic signal sample playback method based on clustering and density assessment disclosed in the embodiments of the present invention can be directly implemented and executed by a hardware processor, or by a combination of hardware and software modules within the processor.

[0151] As a non-volatile computer-readable storage medium, memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules. Memory can include at least one type of storage medium, for example, can include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), programmable read-only memory (Programmable Read Only Memory, PROM), read-only memory (Read Only Memory, ROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), magnetic memory, disk, optical disk, etc. Memory is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory in the embodiment of the present invention can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0152] By designing and programming a processor, the code corresponding to the discrete electromagnetic signal sample playback method based on clustering and density assessment described in the aforementioned embodiment can be embedded into the chip, enabling the chip to execute the steps of the method in the aforementioned embodiment during operation. Designing and programming a processor is well known to those skilled in the art and will not be further described here.

[0153] Based on the same inventive concept, an embodiment of the present invention further provides a storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer executes the discrete electromagnetic signal sample playback method based on clustering and density evaluation discussed above.

[0154] In some optional embodiments, the present invention also provides various aspects of a discrete electromagnetic signal sample playback method based on clustering and density assessment, which can also be implemented in the form of a program product, which includes program code. When the program product is run on an apparatus, the program code is used to enable the control device to execute the steps of a discrete electromagnetic signal sample playback method based on clustering and density assessment according to various exemplary embodiments of the present invention described above in this specification.

[0155] It should be noted that although several units or subunits of the device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of a unit described above can be further divided into multiple units to be embodied. In addition, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0156] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0157] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a server, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0158] Program code for performing the operations of the present invention may be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0159] Where a remote computing device is involved, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0160] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0162] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A discrete electromagnetic signal sample playback method based on clustering and density evaluation, characterized in that: include: Preprocessing samples of discrete electromagnetic signals; Perform clustering and importance evaluation on the preprocessed samples to obtain an ordered playback sample set; Perform density evaluation on samples in the ordered playback sample set and calculate the comprehensive importance score; Screen the ordered playback sample set based on the comprehensive importance score; Resample the screening results under density constraints to generate a playback sample set; The clustering and importance evaluation of the preprocessed samples includes: Use K-Means strategy to initialize cluster centers to ensure that cluster centers are evenly distributed; Iteratively perform sample attribution updates and cluster center recalculations until the cluster center position changes within the required range or the maximum number of iterations is reached, and obtain the cluster label of each sample and the number of samples in each cluster; By using histogram and joint density distribution methods to evaluate the importance of sample types, clustered compact samples and edge samples are selected from the cluster labels of samples. The clustered compact samples are placed at the front and the edge samples are placed at the back to obtain an ordered playback sample set.

2. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 1, characterized in that: The preprocessing of the discrete electromagnetic signal samples refers to normalizing the feature vectors of each category in the original discrete electromagnetic signal samples.

3. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 2, characterized in that: The normalization process includes: A sample of a discrete electromagnetic signal represents a set of pulse-to-pulse parameters, including pulse repetition interval, carrier frequency, and pulse width; The pulse repetition interval, carrier frequency and pulse width are normalized using their respective minimum and maximum values.

4. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 3, characterized in that: The minimum and maximum values ​​are determined based on the data distribution of the dataset, including: Get the minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width in the data set; Expand the minimum and maximum values ​​of the pulse repetition interval, carrier frequency, or pulse width in the data set according to actual application requirements; The minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width in the sample frequency band and the expanded data set are used to calculate the corresponding minimum and maximum values ​​of the pulse repetition interval, carrier frequency or pulse width.

5. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 1, characterized in that: The density evaluation of samples in the ordered playback sample set and calculation of the comprehensive importance score include: In each cluster, the Gaussian kernel function is used to estimate the sample density, and the sample density value of the kernel density function of the clustered compact samples and edge samples in the ordered playback sample set is obtained; Combined with the sample density value and the distance from the sample to the cluster center, the comprehensive importance score is calculated according to the preset weight.

6. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 5, characterized in that: The screening of the ordered playback sample set according to the comprehensive importance score includes: Arrange the samples in the ordered playback sample set in descending order according to the comprehensive importance score, and extract the first proportion of samples after descending order as the high-importance sample set; After filtering out the high-importance sample set, the remaining samples are sorted in ascending order of comprehensive importance scores and the second-highest proportion of samples are extracted as the edge sample set; The high-importance sample set and the marginal sample set are merged in sequence to form a candidate replay list.

7. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 6, characterized in that: The screening results are resampled under density constraints, including: Resample the candidate replay list: undersample the high-importance sample set and oversample the marginal sample set; Merge the resampling results of all categories to obtain the playback dataset and ensure that the total number of samples in each category meets the overall capacity constraint; The replay dataset is sorted again by the comprehensive importance score and the final replay sample set is output.

8. The discrete electromagnetic signal sample playback method based on clustering and density evaluation according to claim 7, characterized in that: The ratio of undersampling to oversampling can be dynamically adjusted according to system storage and computing resources.

9. A discrete electromagnetic signal sample playback device based on clustering and density evaluation, characterized in that: include: A first processing unit, configured to pre-process samples of discrete electromagnetic signals; The second processing unit is used to cluster and evaluate the importance of the preprocessed samples to obtain an ordered playback sample set; The third processing unit is used to perform density evaluation on samples in the ordered playback sample set and calculate a comprehensive importance score; a fourth processing unit, configured to screen the ordered playback sample set according to the comprehensive importance score; a fifth processing unit, configured to resample the screening results under density constraints to generate a playback sample set; The clustering and importance evaluation of the preprocessed samples includes: Use K-Means strategy to initialize cluster centers to ensure that cluster centers are evenly distributed; Iteratively perform sample attribution updates and cluster center recalculations until the cluster center position changes within the required range or the maximum number of iterations is reached, and obtain the cluster label of each sample and the number of samples in each cluster; By using histogram and joint density distribution methods to evaluate the importance of sample types, clustered compact samples and edge samples are selected from the cluster labels of samples. The clustered compact samples are placed at the front and the edge samples are placed at the back to obtain an ordered playback sample set.

Images