An electromagnetic signal recognition method and system based on energy entropy template matching

Abstract

The application provides an electromagnetic signal recognition method and system based on energy entropy template matching, relates to the technical field of electromagnetic signal processing, and solves the technical problems that the existing technology has high calculation complexity of electromagnetic signal feature extraction, low template matching efficiency and difficulty in realizing accurate positioning of long-time signals. The method specifically comprises the following steps: acquiring time domain data of an effective duration stage of a known category electromagnetic signal; generating a time domain energy entropy sequence of the known category electromagnetic signal based on the time domain data, as a signal template; storing the signal template and constructing a signal template library; acquiring a global time domain energy entropy sequence of a to-be-detected electromagnetic signal; determining a target signal template from the signal template library; and determining a recognition result of the to-be-detected electromagnetic signal by sliding window correlation matching based on the global time domain energy entropy sequence and the target signal template. The application is used for recognizing electromagnetic signals.

Description

Technical Field

[0001] This application relates to the field of electromagnetic signal processing technology, and in particular to an electromagnetic signal recognition method and system based on energy entropy template matching. Background Technology

[0002] Electromagnetic signal identification is a key technology in fields such as radio monitoring, spectrum management, and communication reconnaissance, and is of great significance for ensuring electromagnetic spectrum security. Existing electromagnetic signal identification methods mainly rely on signal spectrum analysis, modulation identification, or deep learning classification. For example, spectrum template-based identification methods require FFT frequency domain transformation of the signal, resulting in high computational overhead. Furthermore, the spectrum template is sensitive to changes in the absolute power and noise floor of the signal, limiting its robustness in practical non-cooperative reception scenarios. Deep learning-based identification methods require a large number of labeled samples for model training, demanding high computational resources. Moreover, the model needs to be retrained or fine-tuned when the signal database expands, resulting in poor deployment flexibility. In addition, existing matching methods mostly rely on static feature distance calculations, making it difficult to determine the precise location of target signals in long-term signals. They also lack lightweight modeling and fast matching mechanisms for long-term signals, leading to low identification efficiency and inability to meet real-time monitoring requirements when processing large-scale signal databases or long-term signals. Therefore, existing technologies suffer from high computational complexity in electromagnetic signal feature extraction, low template matching efficiency, and difficulty in achieving precise long-term signal localization. Summary of the Invention

[0003] This application provides an electromagnetic signal recognition method and system based on energy entropy template matching, which solves the technical problems of high computational complexity of electromagnetic signal feature extraction, low template matching efficiency, and difficulty in achieving accurate long-term signal positioning in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for electromagnetic signal recognition based on energy entropy template matching is provided, comprising: acquiring time-domain data of the effective duration of a known category of electromagnetic signal; generating a time-domain energy entropy sequence of the known category of electromagnetic signal based on the time-domain data, as a signal template; storing the signal template to construct a signal template library; acquiring a global time-domain energy entropy sequence of the electromagnetic signal to be detected; determining a target signal template from the signal template library; and determining the recognition result of the electromagnetic signal to be detected by sliding window correlation matching based on the global time-domain energy entropy sequence and the target signal template.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, a time-domain energy entropy sequence of electromagnetic signals of known categories is generated based on time-domain data, including: performing sliding window framing on the time-domain data to obtain multiple time-domain signal frames; calculating the energy value of each time-domain signal frame; determining the energy entropy value of each time-domain signal frame based on the energy value of each time-domain signal frame; and arranging the energy entropy values ​​of each time-domain signal frame in chronological order to form a time-domain energy entropy sequence.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, the energy entropy value of each time-domain signal frame is determined by calculating the following formula:

[0007] in, The energy entropy value. , Let i be the energy value of the i-th time-domain signal frame. ε represents the total number of frames in the valid signal phase, j represents the index of the valid signal phase, and ε represents the preset zero-prevention factor.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the identification result of the electromagnetic signal to be detected is determined by sliding window correlation matching, including: setting a sliding window, the length of which is equal to the length of the target signal template; sliding the sliding window point by point on the global time-domain energy entropy sequence; for each sliding position, calculating the Pearson correlation coefficient between the sequence segment within the sliding window and the target signal template; when the Pearson correlation coefficient exceeds a preset detection threshold, determining that a signal matching the target signal template has been detected at the corresponding sliding position.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the sliding window moves by a step of one energy entropy value each time.

[0010] In conjunction with the first aspect above, in one possible implementation, before determining the identification result of the electromagnetic signal to be detected, the method further includes: loading multiple signal templates from the signal template library into the video memory of the graphics processor; transferring the global temporal energy entropy sequence into the video memory of the graphics processor; and simultaneously calculating the sliding window correlation coefficient between the global temporal energy entropy sequence and each signal template through the multi-threaded parallel computing architecture of the graphics processor.

[0011] In conjunction with the first aspect above, in one possible implementation, after determining the identification result of the electromagnetic signal to be detected, the method further includes: when a matching signal is detected, recording the peak value of the correlation coefficient when the matching is successful; when the peak value of the correlation coefficient is within a preset high threshold range, updating the corresponding signal template in the signal template library based on the time-domain energy entropy sequence within the current sliding window; when the peak value of the correlation coefficient is within a preset low threshold range and higher than a preset detection threshold, temporarily storing the time-domain energy entropy sequence within the current sliding window as a candidate new template, and adding it as a new signal template to the signal template library when the cumulative occurrence number of candidate new templates exceeds a preset number threshold.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, before determining the target signal template from the signal template library, the method further includes: downsampling each signal template in the signal template library to construct a coarse-grained template index layer; downsampling the global time-domain energy entropy sequence by the same factor to obtain a coarse-grained sequence to be detected; performing a sliding window correlation screening between the coarse-grained sequence to be detected and the coarse-grained template index layer to select a set of candidate templates whose correlation coefficient exceeds the screening threshold; and performing sliding window correlation matching only on the signal templates in the candidate template set.

[0013] In conjunction with the first aspect above, in one possible implementation, obtaining time-domain data of the effective duration phase of a known category electromagnetic signal includes: performing effective signal detection on the known category electromagnetic signal to determine the start and end points; extracting time-domain data between the start and end points of the known category electromagnetic signal as the time-domain data of the effective duration phase; wherein, the effective signal detection employs any one of the energy threshold method, the double sliding window method, or the short-time energy zero-crossing rate method.

[0014] Secondly, an electromagnetic signal recognition system based on energy entropy template matching is provided, comprising: a signal acquisition module for acquiring time-domain data of the effective duration of a known category of electromagnetic signal; an energy entropy sequence calculation module for generating a time-domain energy entropy sequence of the known category of electromagnetic signal based on the time-domain data, serving as a signal template; a template library storage module for storing signal templates and constructing a signal template library; a global sequence acquisition module for acquiring a global time-domain energy entropy sequence of the electromagnetic signal to be detected; a target template determination module for determining a target signal template from the signal template library; and a decision output module for determining the recognition result of the electromagnetic signal to be detected based on the global time-domain energy entropy sequence and the target signal template through sliding window correlation matching.

[0015] This invention constructs a lightweight template by directly calculating the energy entropy sequence in the time domain, avoiding complex frequency domain transformations and reducing computational complexity. At the same time, it uses sliding window correlation matching to achieve accurate positioning of target signals in long-term signals, solving the technical problems of high computational complexity of electromagnetic signal feature extraction, low template matching efficiency, and difficulty in achieving accurate positioning of long-term signals in existing technologies.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 2 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 3 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 4 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 5 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 6 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 7 A flowchart illustrating another electromagnetic signal identification method based on energy entropy template matching provided in this application embodiment; Figure 8 This is a system architecture diagram of an electromagnetic signal recognition system based on energy entropy template matching, provided for an embodiment of this application. Detailed Implementation

[0018] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0019] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0020] Example 1: like Figure 1 As shown, this embodiment provides an electromagnetic signal identification method based on energy entropy template matching. The method includes a template construction stage and a signal identification and matching stage.

[0021] The template construction phase specifically includes S101 to S103: S101. Obtain time-domain data of the effective duration of a known type of electromagnetic signal.

[0022] Among them, known category electromagnetic signals refer to electromagnetic signals whose signal type, modulation method or purpose has been determined through prior knowledge or previous monitoring. The effective duration period refers to the period when the signal actually exists and the signal-to-noise ratio meets the processing requirements, excluding the silent period before the signal appears and the noise period after the signal disappears.

[0023] In this embodiment, the original radio frequency (RF) or intermediate frequency (IF) signal can be acquired by connecting a radio receiver or signal acquisition device, and then converted from analog to digital to obtain a discrete time-domain sampling point sequence. Subsequently, using an energy detection algorithm or a spectrum sensing algorithm, the start and end times of the signal are located in the long-term observation data, and the data segment between them is extracted as the time-domain data of the effective duration. It should be understood that the method of acquiring time-domain data is not limited to direct acquisition; it can also be historical data files read from a storage medium.

[0024] Based on the above steps, by removing invalid noise data segments and retaining only the effective part of the signal, the computational load of subsequent feature extraction can be significantly reduced, and the signal-to-noise ratio of the constructed template can be improved, laying the foundation for subsequent high-precision matching.

[0025] S102. Based on time-domain data, generate time-domain energy entropy sequences of electromagnetic signals of known categories as signal templates.

[0026] Among them, the time-domain energy entropy sequence refers to the one-dimensional feature vector formed by dividing time-domain data into multiple short frames in chronological order, calculating the energy entropy value of the signal in each frame, and arranging these entropy values ​​in chronological order. Energy entropy is the application of the concept of information entropy to the distribution of signal energy, used to measure the complexity and uncertainty of the distribution of signal energy in the time domain.

[0027] In this embodiment, the time-domain data acquired in S101 is processed by framing, with each frame containing a certain number of sampling points. The energy value of the signal in each frame is calculated, and the corresponding energy entropy value is calculated based on this energy value. By concatenating the energy entropy values ​​of all frames in chronological order, the time-domain energy entropy sequence of the known category of electromagnetic signals can be obtained. This sequence serves as a feature fingerprint, representing the energy distribution pattern of this type of signal in the time domain, and is therefore stored as a signal template.

[0028] It should be understood that the time-domain energy entropy sequence is one-dimensional data. Compared with two-dimensional features such as spectrograms or time-frequency graphs, its data volume is greatly compressed, and it has the characteristics of being lightweight.

[0029] It should be noted that the time-domain energy entropy sequence is calculated directly in the time domain, without the need for Fast Fourier Transform (FFT) or other complex time-frequency transformations. This results in low computational complexity, making it suitable for deployment in embedded devices or real-time processing systems. Furthermore, the entropy value is insensitive to changes in the absolute power of the signal, exhibiting good noise immunity and robustness.

[0030] As an example, for a 10-millisecond linear frequency modulated (LFM) signal time-domain data, it is divided into 10 frames with a window length of 1 millisecond. The energy entropy value of each frame is calculated, resulting in a sequence containing 10 values. This sequence is the template for the LFM signal.

[0031] Based on the above steps, by generating a time-domain energy entropy sequence as a template, lightweight modeling of signal features is achieved. This not only preserves the key time-domain features of the signal but also greatly compresses the data dimension, solving the problems of large storage space and high computational overhead of spectrum templates in existing technologies.

[0032] S103. Store signal templates and build a signal template library.

[0033] The signal template library refers to a database or data structure used to store various known signal templates, supporting template writing, reading, retrieval, and update operations.

[0034] In this embodiment, the signal template generated in S102 is associated with the category label of the signal (such as "UAV image transmission", "radar pulse", etc.) and stored in a non-volatile storage medium, such as a hard disk, solid-state drive, or cloud storage server. As the number of known signals increases, the template library is continuously expanded to form a feature library containing multiple signal types.

[0035] As an example, the generated drone image transmission signal template is stored in position 1 of the database, the radar pulse signal template is stored in position 2, and so on, to build a template library containing dozens or even hundreds of signal types.

[0036] Based on the above steps, by constructing a signal template library, a standard reference system is provided for subsequent matching and recognition, enabling the system to determine the signal category by comparing the similarity between the signal to be detected and the templates in the library.

[0037] The signal identification and matching stage specifically includes S104 to S106: S104. Obtain the global time-domain energy entropy sequence of the electromagnetic signal to be detected.

[0038] The electromagnetic signal to be detected refers to an electromagnetic signal of unknown type received in the actual monitoring environment. The global time-domain energy entropy sequence refers to the energy entropy sequence obtained by processing the long-term observed signal data for the entire time period, which may include background noise, silent periods, and target signal periods.

[0039] In this embodiment, the receiver continuously monitors a specific frequency band and acquires long-term continuous time-domain data. Using the same parameter settings as in S102 (such as frame length and frame shift), the long-term data is processed frame by frame to calculate the energy entropy value of each frame, and arranged in chronological order to form a global time-domain energy entropy sequence.

[0040] As an example, a monitoring station continuously monitors a certain frequency band for 24 hours, converting the time-domain data collected throughout the day into an energy entropy sequence. This sequence may be millions of points long and contains the characteristic trajectories of all signal activities within that frequency band.

[0041] Based on the above steps, feature extraction of monitoring data was achieved by converting the long-term signal to be detected into a global time-domain energy entropy sequence, providing a data foundation for locating target signals in long-term data.

[0042] S105. Determine the target signal template from the signal template library.

[0043] The target signal template refers to one or more signal templates that need to be compared with the signal to be detected in the current recognition task.

[0044] In this embodiment, a specific template can be selected as the target signal template from the signal template library according to the needs of the monitoring task. For example, if the current task only needs to monitor drone signals, only drone-related templates in the library are selected as targets; if a full library search is required, all templates in the library are used as target signal templates. It should be understood that the determination of the target signal template can be preset or dynamically adjusted based on the spectrum sensing results.

[0045] It should be noted that the process of determining the target signal template is a prerequisite for matching and recognition. By filtering the target template, unnecessary computation can be reduced and recognition efficiency can be improved.

[0046] As an example, in a radio monitoring task, if the target of interest is set as "frequency hopping signal", the system will automatically extract all templates marked as "frequency hopping signal" from the template library as the target signal template.

[0047] Based on the above steps, by determining the target signal template, the scope of the matching search is limited, making the subsequent matching process more targeted.

[0048] S106. Based on the global time-domain energy entropy sequence and the target signal template, the identification result of the electromagnetic signal to be detected is determined by sliding window correlation matching.

[0049] Among them, sliding window correlation matching refers to setting a sliding window with the same length as the target signal template, moving it point by point on the global time domain energy entropy sequence, calculating the correlation coefficient between the sequence segment within the window and the template, and determining whether a matching signal exists based on the magnitude of the correlation coefficient.

[0050] In this embodiment, the length of the sliding window is set to be equal to the length of the target signal template. The sliding window is placed at the beginning of the global sequence, and the correlation coefficient between the sequence within the window and the template is calculated. Then, the sliding window is slid backward point by point, and the correlation coefficient is calculated again. This process is repeated until the sliding window moves to the end of the global sequence. The correlation coefficients at all positions are recorded. When the correlation coefficient at a certain position exceeds a preset detection threshold, it is determined that a signal matching the target template has been detected at that position, and the identification result, including the signal category and the time of occurrence, is output. It should be understood that the correlation coefficient can be calculated using the Pearson correlation coefficient or other similarity measurement methods.

[0051] As an example, the target template has a length of 100 entropy points, and the global sequence has a length of 10,000 points. The sliding window starts from point 1 and moves one point at a time, calculating the correlation coefficient 9,901 times. If the correlation coefficient reaches its peak at point 3000 and exceeds the threshold, it is determined that the target signal appeared at the time corresponding to point 3000 in the global sequence.

[0052] Based on the above steps, by using sliding window correlation matching, automatic search, precise positioning and category identification of target signals in long-term monitoring data are realized, effectively solving the problem of difficulty in accurately locating target signals in long-term signals in existing technologies.

[0053] Based on the above technical solution, this embodiment avoids complex frequency domain transformation by constructing a lightweight time-domain energy entropy template library, thereby reducing computational complexity and storage requirements; and achieves accurate positioning and identification of target signals in long-term monitoring data through a sliding window correlation matching mechanism, improving the accuracy and practicality of identification.

[0054] Example 2: This embodiment, based on Embodiment 1, provides a detailed explanation of the specific process for generating time-domain energy entropy sequences of electromagnetic signals of known categories. For example... Figure 2 As shown, specifically including S201 to S204: S201. Perform sliding window framing on the time-domain data to obtain multiple time-domain signal frames.

[0055] Sliding window framing refers to using a fixed-length window to slide along the time axis to divide a continuous long time-domain signal into several short-time signal segments, each segment being called a time-domain signal frame.

[0056] In this embodiment, setting the window length and step size of the sliding window are key parameters. The window length determines the time span of each frame of signal, and is usually set to a length that can cover the basic period of the signal characteristics. For example, for communication signals, the window length can be set to 1 millisecond to 10 milliseconds. The step size determines the degree of overlap between frames. If the step size is smaller than the window length, there is overlap. The higher the overlap rate, the higher the temporal resolution of the feature sequence, but the computational load also increases accordingly. Specifically, the window length can be set to 5 milliseconds and the step size to 2.5 milliseconds, i.e., a 50% overlap rate, to segment the time-domain data acquired in S101 to obtain a continuous time-domain signal frame sequence.

[0057] It should be noted that by using sliding window framing, the non-stationary time-varying signal is transformed into a series of quasi-stationary short-time signals, so that the energy entropy value calculated subsequently can reflect the characteristics of the signal in a local time period, thereby constructing a feature sequence that changes over time.

[0058] As an example, for a UAV image transmission signal with a sampling rate of 100MHz, the window length is set to 500 sampling points (corresponding to 5 microseconds), and the step size is 250 sampling points. The time domain data is then segmented to obtain several time domain signal frames.

[0059] Based on the above steps, fine segmentation of the signal's temporal features was achieved through sliding window framing, laying the foundation for subsequent extraction of local energy features.

[0060] S202. Calculate the energy value of each time-domain signal frame.

[0061] The energy value refers to the sum of the squares of the amplitudes of all sampling points within a signal frame, reflecting the signal strength or power during that time period.

[0062] In this embodiment, for each time-domain signal frame, the squared amplitudes of all its sample points are calculated, and the summation yields the total energy of the frame. Specifically, assuming the i-th time-domain signal frame contains M sample points, its energy value... It is equal to the sum of the squares of the amplitudes of the M sampling points within the frame.

[0063] It should be noted that calculating energy values ​​is the basis for constructing energy distribution, and the differences in energy values ​​between different signal frames reflect the fluctuation of signal power over time.

[0064] Based on the above steps, the intensity characteristics of the signal in each local time period were quantified by calculating the energy value of each frame.

[0065] S203. Based on the energy values ​​of each time-domain signal frame, determine the energy entropy value of each time-domain signal frame.

[0066] Among them, the energy entropy value is an index that measures the uniformity of signal energy distribution using the principle of information entropy. The larger the entropy value, the more uniform the energy distribution and the more complex the signal; the smaller the entropy value, the more concentrated the energy distribution and the more regular the signal.

[0067] In this embodiment of the application, the energy values ​​of each frame are first normalized, and the proportion of energy of each frame to the total energy is calculated to obtain the energy probability density; then the information entropy is calculated based on the probability density.

[0068] Optionally, the energy entropy value of each time-domain signal frame is determined by calculating the following formula:

[0069] in, The energy entropy value, , Let i be the energy value of the i-th time-domain signal frame. ε represents the total number of frames in the valid signal phase, j represents the index of the valid signal phase, and ε represents the preset zero-prevention factor.

[0070] Specifically, in the formula It represents the proportion of the signal energy of the i-th frame in the total energy, and can be regarded as the "probability" of the occurrence of that frame. This is a measure of the uncertainty of the probability distribution. Introducing... (For example, the value is 10) -9 As a zero-prevention factor, it is used to prevent the energy of a frame from being extremely low or zero. Approaching 0 causes a numerical overflow error in the log(0) operation, thus ensuring the numerical stability of the algorithm under various signal conditions. It should be understood that the base of the logarithm can be 2, e, or 10 without affecting the relative trend of the entropy value.

[0071] It should be noted that energy entropy focuses on the relative shape of energy distribution, rather than its absolute magnitude. Therefore, even if the overall signal power attenuates due to changes in transmission distance, its energy distribution remains essentially unchanged, and the energy entropy value remains stable. This gives this characteristic extremely strong robustness against power fluctuations.

[0072] As an example, for a linear frequency modulated (LFM) signal, its frequency changes linearly with time, and the energy is relatively uniformly distributed in the time domain, resulting in a stable high-value energy entropy sequence. In contrast, for a burst pulse signal, the energy is concentrated in a short period, and the energy entropy sequence exhibits a significant spike characteristic. Different signals have drastically different entropy sequence shapes, which is the physical basis for their use as identification features.

[0073] Based on the above steps, by calculating the energy entropy value, the time-domain energy distribution of the signal is transformed into a physically meaningful feature scalar, and this feature is insensitive to power changes, thus improving the robustness of recognition.

[0074] S204. Arrange the energy entropy values ​​of each time-domain signal frame in chronological order to form a time-domain energy entropy sequence.

[0075] Among them, the temporal energy entropy sequence is a one-dimensional vector composed of the energy entropy values ​​of each frame in chronological order.

[0076] In this embodiment, the energy entropy values ​​of each frame calculated in S203 are concatenated according to the temporal order of the corresponding time-domain signal frames to construct a one-dimensional time-domain energy entropy sequence. The length of this sequence is equal to the total number of frames N in the effective signal phase.

[0077] It should be noted that compared with two-dimensional features such as spectrograms or time-frequency graphs, one-dimensional time-domain energy entropy sequence data is extremely small in size and has low storage and transmission overhead, making it very suitable for building large-scale lightweight template libraries.

[0078] As an example, the calculated N energy entropy values ​​are arranged in order to form a structure like [ , , ,..., The sequence of [] serves as the signal template for the known category of electromagnetic signals.

[0079] Based on the above steps, by constructing a time-domain energy entropy sequence, a lightweight and robust model of signal features is achieved, providing efficient data support for subsequent template matching.

[0080] Based on the above technical solution, this embodiment elaborates on the algorithm for generating time-domain energy entropy sequences. Through sliding window framing, energy calculation, entropy calculation and sequence construction, a one-dimensional feature that can characterize the time-domain energy distribution law of the signal and is insensitive to power fluctuations is extracted. This not only explains the physical mechanism of the algorithm, but also ensures the numerical stability of the algorithm by introducing a zero-prevention factor, providing a reliable feature basis for signal recognition.

[0081] Example 3: This embodiment, based on Embodiment 1, provides a detailed explanation of the specific process for determining the identification result of the electromagnetic signal to be detected through sliding window correlation matching. For example... Figure 3 As shown, specifically including S301 to S304: S301. Set a sliding window, the length of which is equal to the length of the target signal template.

[0082] The sliding window refers to a virtual window that extracts a data segment from the global time-domain energy entropy sequence, and its length determines the amount of data involved in each matching calculation.

[0083] In this embodiment, the length L of the target signal template is read, and the window length of the sliding window is set to L. This setting ensures that the sequence segment extracted by the sliding window is perfectly aligned with the template sequence on the time scale, thus making subsequent correlation coefficient calculations comparable. If the window length is inconsistent with the template length, effective similarity measurement cannot be performed.

[0084] It should be understood that a sliding window is essentially a data extractor of length L, which continuously extracts local segments from the global sequence as it moves along the time axis.

[0085] It should be noted that setting the sliding window length to match the template length is a prerequisite for achieving precise positioning. Only when the window length matches the template length can the peak value of the correlation coefficient accurately indicate the starting position of the target signal in the global sequence.

[0086] As an example, if the target signal template contains 100 energy entropy points, the window length of the sliding window is also set to 100 points.

[0087] Based on the above steps, by setting the length of the matching sliding window, the scale consistency of the matching features is ensured, providing a standardized data foundation for subsequent related calculations.

[0088] S302. Slide the sliding window point by point on the global time-domain energy entropy sequence.

[0089] Point-by-point sliding refers to the sliding window moving in steps of one energy entropy value point each time. That is, when the sliding window moves from the current position to the next position, it only shifts backward by one data point.

[0090] In this embodiment, the sliding window is first placed at the beginning of the global temporal energy entropy sequence (e.g., the 1st to the Lth point), and the correlation coefficient at that position is calculated. Then, the sliding window moves one point forward, covering the 2nd to the (L+1th)th points, and the correlation coefficient is calculated again. This process is repeated until the sliding window reaches the end of the global sequence.

[0091] It should be understood that a point-by-point sliding motion with a step size of 1 ensures that no possible signal starting position is missed, achieving the highest accuracy in the search. Although increasing the step size can reduce the amount of computation, it may lead to a decrease in positioning accuracy or the failure to detect short-term signals.

[0092] It should be noted that although point-by-point sliding involves relatively large computational load, it can capture subtle changes in the correlation coefficient, thereby achieving precise locking of the start and end times of the signal to be detected. This is a key step that distinguishes it from coarse-grained detection.

[0093] Based on the above steps, a point-by-point sliding method was used to achieve full-time scanning of long-term monitoring data, ensuring the accuracy and completeness of signal positioning.

[0094] S303. For each sliding position, calculate the Pearson correlation coefficient between the sequence segment within the sliding window and the target signal template.

[0095] The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two variables. Its value ranges from -1 to 1, and the larger the absolute value, the stronger the correlation.

[0096] In this embodiment, for each sliding position, L energy entropy values ​​within the sliding window are extracted to form a vector X, and L energy entropy values ​​of the target signal template are combined to form a vector Y. The Pearson correlation coefficient r between vector X and vector Y is calculated. The specific calculation formula is as follows:

[0097] in, Let i be the i-th value of the sequence within the sliding window. For the i-th value of the template sequence, , , where are the mean values ​​of the corresponding sequences, and L is the length of the template sequence. The Pearson correlation coefficient has a normalizing property, meaning it is insensitive to changes in the amplitude of the sequence. This means that even if the power of the signal to be detected differs from the template signal due to variations in transmission distance, the correlation coefficient will still be high as long as their energy distribution patterns are similar. This is the core reason why this embodiment chooses the Pearson correlation coefficient instead of Euclidean distance as the similarity measure.

[0098] It should be noted that traditional Euclidean distance metrics are sensitive to the absolute amplitude of the signal. When the signal power attenuates, the distance value increases significantly, leading to misjudgments. In contrast, the Pearson correlation coefficient, through normalization by subtracting the mean and dividing by the standard deviation, eliminates the influence of absolute power, focusing only on the morphological similarity of the sequence waveform. This greatly improves the robustness of the recognition algorithm in non-cooperative reception scenarios.

[0099] As an example, suppose the entropy of the template sequence ranges from [0.5, 0.8], while the entropy of the signal to be detected decreases to the range of [0.2, 0.5] due to power attenuation, but the waveform trends are completely consistent. In this case, the Euclidean distance will be relatively large, while the Pearson correlation coefficient will still be close to 1, indicating that the two are highly correlated.

[0100] Based on the above steps, by calculating the Pearson correlation coefficient, a precise measurement of signal morphological similarity is achieved, effectively overcoming the impact of signal power fluctuations on recognition accuracy.

[0101] S304. When the Pearson correlation coefficient exceeds the preset detection threshold, it is determined that a signal matching the target signal template is detected at the corresponding sliding position.

[0102] The preset detection threshold is a critical value used to determine whether a match is valid. It is usually set according to experimental statistics or false alarm rate requirements, such as 0.85 or 0.9.

[0103] In this embodiment, the correlation coefficient calculated at each sliding position is monitored in real time. When the correlation coefficient curve shows a peak value exceeding a preset detection threshold, it is determined that a target signal exists at that position. At this time, the starting position of the sliding window is the starting time of the signal to be detected in the global sequence, and the ending position of the sliding window is the ending time of the signal. It should be understood that the appearance of the correlation coefficient peak value means that at that moment, the energy entropy sequence of the signal to be detected and the template sequence have achieved optimal matching.

[0104] As an example, the detection threshold is set to 0.9. When the sliding window moves to the 3000th point, the calculated correlation coefficient is 0.95, which exceeds the threshold. Therefore, it is determined that the target signal was detected at the time corresponding to the 3000th point, and this time is recorded as the signal starting point.

[0105] Based on the above steps, the automatic confirmation and location output of the matching results are realized through threshold determination, thus completing the final closed loop from feature matching to signal recognition.

[0106] Based on the above technical solution, this embodiment achieves precise localization and highly robust recognition of target signals in long-term signals by setting a sliding window of the same length as the template for point-by-point search and using the Pearson correlation coefficient as a similarity metric. This method effectively overcomes the impact of signal power fluctuations on the recognition results and significantly improves the recognition accuracy in non-cooperative reception scenarios.

[0107] Example 4: This embodiment, based on Embodiment 1, provides a detailed explanation of the specific process of parallel acceleration using a graphics processing unit (GPU). For example... Figure 4 As shown, specifically including S401 to S403: S401. Load multiple signal templates from the signal template library into the graphics processor's video memory.

[0108] The Graphics Processing Unit (GPU) is a processor specifically designed for handling graphics and parallel computing. It has thousands of computing cores and high-bandwidth video memory, making it ideal for handling large-scale data-parallel computing tasks. Video memory is dedicated to the GPU, and its read / write speed is much faster than the computer's main memory.

[0109] In this embodiment of the application, during the initialization phase or after the template library is updated, the signal template library stored in the hard disk or main memory is read in batches and transmitted to the GPU's video memory via the PCIe bus for permanent storage.

[0110] It should be understood that preloading template data into video memory avoids the repeated data transfer overhead during each matching, allowing subsequent calculations to fully utilize the high bandwidth of video memory and eliminate data transfer bottlenecks.

[0111] As an example, the system includes a template library containing 1,000 signal templates, each template being 100 floating-point numbers in length. At startup, the system copies all 1,000 templates to the GPU memory for subsequent real-time matching and calling.

[0112] It should be noted that traditional CPU serial processing methods, when handling large-scale template libraries, are often unable to meet real-time requirements due to limitations in memory bandwidth and the number of cores. This step, by migrating data to GPU memory, lays the data foundation for subsequent large-scale parallel computing.

[0113] Based on the above steps, by loading the template library into the GPU memory, the data transmission delay in the matching process is eliminated, making full use of the high bandwidth characteristics of the GPU and providing hardware support for the rapid matching of large-scale template libraries.

[0114] S402: Transfer the global temporal energy entropy sequence to the graphics processor's video memory.

[0115] Among them, the global time-domain energy entropy sequence is the input data to be detected. Its length is usually long and continues to increase with the increase of monitoring time.

[0116] In this embodiment of the application, after the system obtains the global time-domain energy entropy sequence of the electromagnetic signal to be detected, it encapsulates it into a data packet and transmits it to the GPU memory through a high-speed interface (such as PCIe 4.0 or higher).

[0117] It should be understood that the transmission process is streaming, meaning that as the received data accumulates, the data is transmitted to the GPU in batches to achieve pipelined data processing, thereby further reducing processing latency.

[0118] As an example, the monitoring station generates 10,000 energy entropy points per second. The system packages the data from 1,000 points into 100-millisecond increments and transmits them to the GPU memory to ensure real-time data processing.

[0119] Based on the above steps, by transferring the global sequence to the GPU memory, hardware-level alignment between the data to be detected and the template data is achieved, which prepares the ground for subsequent parallel related calculations.

[0120] S403 uses a multi-threaded parallel computing architecture of a graphics processor to simultaneously calculate the sliding window correlation coefficient between the global time-domain energy entropy sequence and each signal template.

[0121] Multithreaded parallel computing architecture refers to an architectural pattern that utilizes a large number of computing cores within the GPU to execute multiple computing tasks simultaneously. In GPU programming models, computing tasks are typically divided into multiple thread blocks and thread grids.

[0122] In this embodiment, a parallel strategy of "one thread block corresponding to one signal template" is adopted. Specifically, the GPU launches N thread blocks, where N is the number of templates in the signal template library. Within each thread block, a sufficient number of threads are allocated to calculate the sliding window correlation coefficient between the template and the global sequence. The thread blocks are independent of each other, execute in parallel, and do not interfere with each other. Each thread block is responsible for calculating the correlation coefficient of a template at all sliding positions in the global sequence and storing the results in video memory.

[0123] It should be understood that this parallel strategy makes the matching computation time no longer increase linearly with the increase of the number of templates, but depends on the computation time of a single thread block, thereby decoupling the matching time from the size of the template library.

[0124] As an example, if there are 500 templates in the template library, the GPU can simultaneously launch 500 thread blocks, each independently calculating the sliding window correlation coefficient for its corresponding template. Regardless of whether the template library size is 100 or 1000, as long as GPU resources are sufficient, the overall matching time remains relatively stable, greatly improving recognition efficiency.

[0125] It should be noted that the traditional CPU serial matching method must traverse the templates in the template library one by one for calculation. As the size of the template library increases, the calculation time increases exponentially, making it difficult to meet the requirements of real-time monitoring. This embodiment, however, utilizes the massive parallel capabilities of the GPU to transform the serial "traversal-computation" process into a parallel "simultaneous computation" process, fundamentally solving the problem of slow matching in large-scale template libraries and demonstrating a complete logical chain of "means-function-effect".

[0126] Based on the above steps, the parallel processing of multi-template matching tasks was realized through a GPU multi-threaded parallel computing architecture, which basically decoupled the matching time from the template library size, significantly improved the recognition efficiency in large-scale signal library scenarios, and met the needs of real-time monitoring.

[0127] Based on the above technical solution, this embodiment loads the template library and the data to be detected into the GPU memory and uses a multi-threaded parallel architecture to calculate the sliding window correlation coefficient simultaneously. This fully utilizes the high concurrency computing capability of the GPU, solves the problem of low efficiency of traditional serial computing in large-scale template library scenarios, and achieves a real-time breakthrough in electromagnetic signal recognition.

[0128] Example 5: This embodiment, based on Embodiment 1, provides a detailed explanation of the adaptive update mechanism of the signal template library. In practical application scenarios where the electromagnetic environment changes dynamically, a fixed template library is difficult to adapt to the drift of signal characteristics and the emergence of new signals; therefore, dynamic maintenance of the template library is necessary. Figure 5As shown, specifically including S501 to S503: S501. When a matching signal is detected, record the peak value of the correlation coefficient when the matching is successful.

[0129] The peak correlation coefficient refers to the maximum value of the Pearson correlation coefficient calculated between the sequence within the sliding window and the target signal template during the sliding window correlation matching process. This peak value reflects the similarity between the signal to be detected and the template.

[0130] In this embodiment of the application, when performing the matching process of Embodiment 1 or Embodiment 3, not only is it determined whether the correlation coefficient exceeds the detection threshold to confirm the presence of a signal, but the specific value of the correlation coefficient at the moment of successful matching is also recorded. It should be understood that the magnitude of this peak value directly characterizes the reliability of the current matching result. The higher the peak value, the more closely the current signal matches the template; conversely, the lower the peak value, the more likely the current signal is distorted or belongs to a new signal.

[0131] It should be noted that recording the peak value of the correlation coefficient is the data basis for subsequent template updates or new signal discovery decisions. Only by quantifying the confidence level of the match can a scientific update strategy be formulated.

[0132] Based on the above steps, by recording the matching peak, a quantitative indicator is provided for the dynamic updating of the template library, enabling the system to make different response strategies based on the matching quality.

[0133] S502. When the peak value of the correlation coefficient is in the preset high threshold range, update the corresponding signal template in the signal template library based on the time domain energy entropy sequence within the current sliding window.

[0134] The preset high threshold range refers to a numerical range of the correlation coefficient, usually set in a high value range close to 1, such as [0.95, 1.0], indicating that the current matching result has a very high confidence level, confirming that it is a known signal and the signal quality is good.

[0135] In this embodiment, when the peak value of the recorded correlation coefficient falls into the high threshold range, it indicates that the currently received signal is a known category signal, and the signal characteristics are highly consistent with the template. At this time, the system uses a weighted average strategy to update the corresponding template in the library. Specifically, let the original template sequence be T_old, the sequence in the current sliding window be T_new, and the updated template T_update = α*T_old + (1-α)*T_new, where α is the forgetting factor or weighting coefficient, with a value between 0 and 1. Through this soft update method, the template can slowly absorb the characteristics of new signals, thereby adapting to the slight drift in signal characteristics caused by changes in the channel environment (such as changes in multipath effects or equipment aging). It should be understood that setting the high threshold range avoids the contamination of the template by low-quality matching, ensuring the stability of the template library.

[0136] It should be noted that this adaptive update mechanism solves the problem of declining recognition rate due to environmental changes in the long-term operation of a fixed template library, enabling the template library to self-optimize as the environment evolves and always maintain a high matching accuracy.

[0137] As an example, for a certain type of UAV image transmission signal, as the UAV's flight distance changes, the power and noise floor of the received signal fluctuate, causing a slight deformation in its energy entropy sequence. When the system detects a correlation coefficient of 0.98, it determines it as a high-confidence match and uses the current sequence to perform a weighted update on the templates in the library, gradually adapting the templates to the current channel environment.

[0138] Based on the above steps, by using weighted updates during high-confidence matching, the template library is dynamically adapted to environmental changes, extending its effective lifespan and improving the system's robustness.

[0139] S503. When the peak value of the correlation coefficient is in the preset low threshold range and higher than the preset detection threshold, the time domain energy entropy sequence in the current sliding window is temporarily stored as a candidate new template, and when the cumulative occurrence of the candidate new template exceeds the preset number threshold, it is added to the signal template library as a new signal template.

[0140] The preset low threshold range refers to an intermediate range of the correlation coefficient, such as [0.85, 0.90]. This indicates that although the match is successful, the confidence level is not high, which may be due to distortion of a known signal or an unknown new signal. The preset detection threshold refers to the lowest threshold for determining the existence of a signal.

[0141] In this embodiment, when the peak correlation coefficient is higher than the detection threshold but falls within the lower threshold range, the system determines that the signal is in a "fuzzy zone," meaning it cannot be definitively identified as a known signal for template updating, nor can it be directly ignored. In this case, the system stores the temporal energy entropy sequence within the current sliding window as a "candidate new template" in a temporary buffer. The system counts these candidate templates; if the same or similar candidate templates appear multiple times during subsequent monitoring, and the cumulative occurrence exceeds a preset threshold (e.g., 5 or 10 times), it indicates that the signal characteristics are stable and frequently occurring, and it is highly likely to be a newly emerging unknown signal type. The system automatically adds it to the signal template library and assigns a new category label. It should be understood that the "cumulative occurrence count" condition is introduced to filter out occasional noise interference or transient interference signals, prevent the inclusion of invalid templates, and ensure the purity of the template library.

[0142] Based on the above steps, the automatic discovery and storage of new signals is achieved through a candidate temporary storage and cumulative confirmation mechanism based on low confidence matching, thereby enhancing the system's scalability and its ability to perceive unknown environments.

[0143] Based on the above technical solution, this embodiment uses a dual-threshold judgment logic to divide the matching results into three intervals: high confidence, low confidence, and invalid, corresponding to three processing strategies: template update, new signal discovery, and ignoring, respectively. This mechanism not only solves the problem that a fixed template library is difficult to adapt to environmental changes, but also endows the system with the ability to automatically discover new signals, significantly improving the practicality and intelligence level of the electromagnetic signal recognition system.

[0144] Example 6: This embodiment, based on Embodiment 1, provides a detailed explanation of the multi-resolution pre-screening mechanism for scenarios involving ultra-large-scale signal template libraries. When the number of templates in the signal template library reaches tens or even hundreds of thousands, directly performing high-resolution sliding window correlation matching on all templates would incur a huge computational burden. Therefore, this embodiment introduces a two-level architecture of "coarse screening - fine screening." For example... Figure 6 As shown, specifically including S601 to S604: S601. Downsample each signal template in the signal template library to construct a coarse-grained template index layer.

[0145] Downsampling refers to the process of reducing the number of sampling points in a data sequence to lower the data resolution, aiming to significantly reduce the amount of data while preserving the main contour features of the signal. The coarse-grained template index layer refers to a lightweight index database composed of downsampled, low-resolution templates.

[0146] In this embodiment, the system uses a preset downsampling factor K (e.g., K=10) to downsample each original time-domain energy entropy sequence template stored in the signal template library. Specifically, it can retain one point every K-1 points, or use mean pooling to compress the original template of length L into a coarse-grained template of length L / K. All coarse-grained templates are combined to construct a coarse-grained template index layer, which resides in memory or video memory for fast retrieval. The choice of downsampling factor K needs to be balanced between computational efficiency and initial screening accuracy; a larger K value results in faster computation but fewer detailed features are retained.

[0147] As an example, the original signal template library contains 10,000 templates, each with 1,000 points. The system sets the downsampling factor to 10, compresses each template, and obtains 10,000 coarse-grained templates with a length of 100 points, thus constructing a coarse-grained template index layer.

[0148] Based on the above steps, by constructing a coarse-grained template index layer, lightweight characterization of a large-scale template library was achieved, laying the foundation for subsequent rapid initial screening.

[0149] S602. Downsample the global time-domain energy entropy sequence by the same factor to obtain a coarse-grained detection sequence.

[0150] Among them, the coarse-grained detection sequence refers to the low-resolution sequence obtained by downsampling the global time-domain energy entropy sequence of the signal to be detected according to the same rules as the template library.

[0151] In this embodiment, to ensure consistency in matching scale, the same downsampling factor K and downsampling algorithm as in S601 are used to process the currently acquired global temporal energy entropy sequence. For example, if the template library uses 10x downsampling, the global sequence is also downsampled by 10x. This ensures that the resulting coarse-grained detection sequence is aligned with the coarse-grained template in terms of time scale, guaranteeing the effectiveness of the correlation coefficient calculation.

[0152] It is important to note that maintaining consistency in downsampling parameters is a prerequisite for ensuring the reliability of initial screening results. If the scales are inconsistent, the correlation coefficient will lose its physical meaning, leading to a large number of missed or false alarms in the initial screening results.

[0153] As an example, the currently received global temporal energy entropy sequence has a length of 100,000 points. The system downsamples it by 10 times to obtain a coarse-grained detection sequence with a length of 10,000 points.

[0154] Based on the above steps, by synchronously downsampling the global sequence, the matching of the data to be detected with the template index layer in terms of feature dimension is ensured, providing compliant input data for the initial screening.

[0155] S603. Perform a sliding window correlation screening between the coarse-grained sequence to be detected and the coarse-grained template index layer to select a set of candidate templates whose correlation coefficient exceeds the initial screening threshold.

[0156] In this context, sliding window correlation screening refers to the process of quickly evaluating the similarity between the sequence to be detected and various templates in a low-resolution data space using sliding window correlation calculations. The screening threshold is a threshold value used to determine whether a template should enter the next round of fine matching; it is usually set relatively low to ensure a high recall rate. The candidate template set refers to a small subset of templates that may match the signal to be detected, selected through the screening process.

[0157] In this embodiment, the sliding window correlation matching algorithm described in Embodiment 3 is used to perform calculations in the coarse-grained space. Since the data length is compressed by a factor of K, the computational complexity of the sliding window is correspondingly reduced. Specifically, assuming the original template length is L, and after downsampling it is L / K, the computational cost of a single correlation calculation is approximately 1 / K of the original. The coarse-grained sequence to be detected is matched with each template in the coarse-grained template index layer, and the peak value of the correlation coefficient is recorded. If the peak value of a template exceeds a preset initial screening threshold, the original version (undownsampled) of that template is added to the candidate template set.

[0158] It should be understood that the purpose of the initial screening is to quickly eliminate a large number of obviously irrelevant templates. Therefore, the initial screening threshold should not be too high to avoid missing real signals.

[0159] As an example, in coarse-grained spatial matching, the correlation coefficient between a certain template and the sequence to be detected is 0.6, which exceeds the initial screening threshold of 0.5. Therefore, the template is determined to be a candidate template and its ID is added to the candidate set. On the other hand, the correlation coefficient of another template is only 0.2, so it is directly excluded and will not participate in the subsequent fine matching.

[0160] Based on the above steps, the number of non-target templates is significantly reduced through rapid initial screening at low resolution, which greatly reduces the computational burden of subsequent fine matching.

[0161] S604. Perform sliding window correlation matching only on signal templates in the candidate template set.

[0162] Among them, sliding window correlation matching refers to the precise matching process performed in the original high-resolution data space, i.e., the method described in Example 3.

[0163] In this embodiment, only the templates in the candidate template set generated in S603 are read from the original signal template library to obtain their corresponding high-resolution temporal energy entropy sequences. Then, these candidate templates are matched with the original global temporal energy entropy sequences using a sliding window correlation method to calculate the precise Pearson correlation coefficient, and the final identification result is determined based on a preset detection threshold. After this initial screening, the size of the candidate template set is typically much smaller than the size of the original template library (e.g., reduced from 10,000 to 50), thus significantly reducing the total computational load for fine matching.

[0164] It should be noted that this two-stage matching architecture, which proceeds from coarse to fine, greatly improves recognition efficiency without sacrificing accuracy. While coarse-grained matching sacrifices some detailed information, it is sufficient to distinguish most non-target signals; fine matching ensures the accuracy of the final recognition result, achieving a balance between efficiency and accuracy.

[0165] As an example, assuming the candidate template set contains only 30 templates, performing sliding window matching at the original resolution on only these 30 templates can improve the recognition speed by hundreds of times.

[0166] Based on the above steps, by performing high-precision matching on candidate templates, the system can achieve rapid searching of a large-scale template library while ensuring recognition accuracy, effectively solving the technical problem of low efficiency in processing large-scale signal libraries in existing technologies.

[0167] Based on the above technical solution, this embodiment constructs a coarse-grained index layer for rapid initial screening, followed by fine matching of a small number of candidate templates, thus building a two-level architecture of "coarse screening-fine screening". This mechanism enables the system to significantly reduce the amount of computation when processing ultra-large-scale template libraries, achieving a doubling of computational efficiency, while ensuring that the recognition accuracy is not affected, greatly improving the practicality and scalability of the method.

[0168] Example 7: This embodiment, based on Embodiment 1, provides a detailed explanation of the specific process for acquiring time-domain data of the effective duration of electromagnetic signals of a known category. In actual electromagnetic environment monitoring, the raw data collected by the receiver often contains a large amount of silent noise segments. Directly extracting features from the raw, long data will introduce noise interference and reduce template quality. Therefore, it is necessary to perform effective signal detection and truncation on the raw signal. For example... Figure 7 As shown, specifically including S701 to S702: S701. Perform effective signal detection on electromagnetic signals of known categories to determine the start and end points.

[0169] Effective signal detection refers to the process of distinguishing between periods of signal presence and periods of pure noise within a continuous received data stream. The starting point is the moment when the signal energy begins to be significantly higher than the noise floor, and the ending point is the moment when the signal energy falls back to the noise floor.

[0170] In this embodiment, the received raw radio frequency or intermediate frequency signal is preprocessed. By calculating characteristic parameters such as short-time energy or zero-crossing rate of the signal, corresponding decision thresholds or rules are set to find the boundary where the signal exists. Specifically, effective signal detection can employ any one of the following methods: energy threshold method, double sliding window method, or short-time energy zero-crossing rate method.

[0171] The energy threshold method calculates the instantaneous power or short-term energy of the signal. When the energy value exceeds a preset noise threshold and remains so for a certain duration, the signal is considered to have started; when the energy value falls below the threshold and remains so for a certain duration, the signal is considered to have ended. This method is simple to calculate and suitable for scenarios with a high signal-to-noise ratio.

[0172] For the dual sliding window method, two adjacent sliding windows (such as a front window and a rear window) are set, and the energy ratio within each window is calculated. When the ratio exceeds a preset threshold, the signal is considered to have started; when the ratio falls below the threshold, the signal is considered to have ended. This method is more sensitive to signal boundary detection and can adapt to situations where signal power changes slowly.

[0173] The short-time energy zero-crossing rate method combines the short-time energy and zero-crossing rate of the signal for joint decision-making. Typically, signal segments have higher energy and relatively stable zero-crossing rates, while noise segments have lower energy and randomly fluctuating zero-crossing rates. This dual-feature decision-making effectively reduces the false alarm rate at low signal-to-noise ratios.

[0174] It should be noted that the three methods described above are merely examples. Those skilled in the art can select appropriate detection methods based on the signal-to-noise ratio and signal type of the actual application scenario, or combine multiple methods to improve detection accuracy. Determining accurate start and end points is a prerequisite for constructing a high-quality template, effectively eliminating noise segments before and after the signal and preventing noise features from being mixed into the template.

[0175] Based on the above steps, through effective signal detection, the precise location of the signal's existence period is achieved, invalid noise data is eliminated, and a guarantee is provided for the subsequent extraction of pure signal features.

[0176] S702. Extract the time-domain data between the start and end points of a known type of electromagnetic signal as the time-domain data of the effective duration phase.

[0177] Here, "truncation" refers to the operation of extracting corresponding segments from the original long-term data sequence based on the detected start and end time indices.

[0178] In this embodiment, based on the start and end point indices determined in step S701, all sampling point data within the range of those indices are extracted from the original time-domain data array. The extracted data segment is the time-domain data of the effective duration phase, which contains only the effective components of the target signal, removing the silence noise before and after it.

[0179] It should be noted that the length of the effective time-domain data extracted may vary depending on the signal type. For example, continuous communication signals have a longer data length, while burst pulse signals have a shorter data length. This embodiment does not impose a specific limitation on the data length, as long as it can cover the effective duration of the signal.

[0180] Based on the above steps, high-quality clean signal samples were obtained by extracting effective time-domain data, eliminating the interference of noise segments on feature extraction, thereby significantly improving the accuracy and representativeness of the signal templates constructed subsequently.

[0181] Based on the above technical solutions, this embodiment accurately determines the start and end times of the signal and extracts the time-domain data of the effective duration phase through detection methods such as the energy threshold method, the double sliding window method, or the short-time energy zero-crossing rate method. This preprocessing effectively eliminates noise interference, ensuring that the data source used for signal template construction has a high signal-to-noise ratio, thus laying the foundation for constructing a high-quality, lightweight template.

[0182] Example 8: This embodiment provides an electromagnetic signal recognition system 80 based on energy entropy template matching. This system and the method embodiment belong to the same inventive concept, and their core lies in achieving efficient electromagnetic signal recognition through the collaborative work of hardware and software modules. Specifically, the system includes a signal acquisition module 801, an energy entropy sequence calculation module 802, a template library storage module 803, a global sequence acquisition module 804, a target template determination module 805, and a decision output module 806.

[0183] The signal acquisition module 801 is used to acquire time-domain data of the effective duration of electromagnetic signals of a known category.

[0184] Specifically, the signal acquisition module 801, serving as the system's input, typically comprises an antenna, a radio frequency (RF) front-end, an analog-to-digital converter (ADC), and a preprocessing unit. The antenna receives electromagnetic wave signals from the air. After filtering, amplification, and down-conversion by the RF front-end, the ADC converts the analog signal into a digital time-domain sampling sequence. The preprocessing unit is responsible for preliminary processing of the raw sampled data, such as valid signal detection as described in Example 7, determining the start and end points of the signal, thereby extracting the time-domain data of the valid duration phase.

[0185] The energy entropy sequence calculation module 802 is used to generate a time-domain energy entropy sequence of electromagnetic signals of known categories based on time-domain data, which serves as a signal template.

[0186] Specifically, the energy entropy sequence calculation module 802 is the core of the system's feature extraction, integrating a sliding window framing unit, an energy calculation unit, and an entropy calculation unit. This module receives time-domain data output from the signal acquisition module 801, performs sliding window framing according to the algorithm flow described in Embodiment 2, calculates the energy of each frame, and finally generates a time-domain energy entropy sequence. This module can be implemented by a digital signal processor (DSP), a field-programmable gate array (FPGA), or a general-purpose processor (CPU) running the corresponding software algorithm. Because its calculation process does not require FFT transformation and consumes low hardware resources, it is suitable for integration into embedded monitoring devices.

[0187] Template library storage module 803 is used to store signal templates and build a signal template library.

[0188] Specifically, the template library storage module 803 is responsible for the persistent storage and management of signal feature data. Its physical carrier can be a hard disk, solid-state drive (SSD), or non-volatile memory. At the logical level, this module maintains a signal template database, where each record contains a signal category label, time-domain energy entropy sequence data, and related metadata (such as sampling rate, window length parameters, etc.). This module also provides a data interface that supports writing, reading, updating, and deleting templates, providing standard reference data for subsequent matching and identification.

[0189] The global sequence acquisition module 804 is used to acquire the global time-domain energy entropy sequence of the electromagnetic signal to be detected.

[0190] Specifically, the global sequence acquisition module 804 functions similarly to the aforementioned energy entropy sequence calculation module 802, but its processing object is the long-term monitoring data to be detected. This module receives continuous time-domain data from the monitoring frequency band in real time and continuously calculates energy entropy values ​​using the same parameter settings as in the template construction stage, generating a global time-domain energy entropy sequence. This sequence is then sent to the decision output module 806 for matching processing. It should be understood that, to improve real-time performance, this module typically employs a pipeline operation method, calculating features while receiving data.

[0191] The target template determination module 805 is used to determine the target signal template from the signal template library.

[0192] Specifically, the target template determination module 805 acts as the system's scheduling center, responsible for filtering the matching range according to the monitoring task requirements. Operators can set the target signal type of interest through a human-machine interface. This module retrieves the corresponding template ID from the template library storage module 803 according to the instructions and loads it into memory or video memory for matching. If the task requires a full library search, this module marks all templates in the library as target signal templates. This mechanism effectively reduces invalid matching calculations and improves system response speed.

[0193] The decision output module 806 is used to determine the identification result of the electromagnetic signal to be detected based on the global time-domain energy entropy sequence and the target signal template through sliding window correlation matching.

[0194] Specifically, the decision output module 806 is the core processing unit of the system, responsible for executing the matching and decision logic described in Embodiments 3 to 6. This module internally implements a sliding window correlation matching algorithm to calculate the Pearson correlation coefficient between the global sequence and the target template. When the correlation coefficient exceeds a preset threshold, the module determines that a target signal has been detected and outputs the identification result, including information such as signal type, occurrence time, and peak correlation coefficient. Furthermore, this module can also integrate the GPU parallel acceleration function described in Embodiment 4, enabling rapid parallel matching of a large-scale template library by utilizing GPU hardware resources, thereby meeting real-time monitoring requirements.

[0195] Based on the above technical solution, this embodiment constructs a complete electromagnetic signal recognition system. Each module has a clear division of labor and works collaboratively, transforming the electromagnetic signal recognition method based on energy entropy template matching into specific hardware or software modules. Each module performs its own function, forming a complete signal processing and recognition chain from front-end signal interception to back-end logical decision-making.

[0196] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0197] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. An electromagnetic signal identification method based on energy entropy template matching, characterized in that, include: Obtain time-domain data of the effective duration of electromagnetic signals of known categories; Based on the time-domain data, a time-domain energy entropy sequence of electromagnetic signals of known categories is generated as a signal template; Store the signal templates to build a signal template library; Obtain the global time-domain energy entropy sequence of the electromagnetic signal to be detected; Determine the target signal template from the signal template library; Based on the global time-domain energy entropy sequence and the target signal template, the identification result of the electromagnetic signal to be detected is determined by sliding window correlation matching.

2. The method according to claim 1, characterized in that, The step of generating a time-domain energy entropy sequence for electromagnetic signals of known categories based on the time-domain data includes: The time-domain data is divided into multiple time-domain signal frames by sliding windowing. Calculate the energy value of each of the time-domain signal frames; Based on the energy value of each time-domain signal frame, determine the energy entropy value of each time-domain signal frame; The energy entropy values ​​of each time-domain signal frame are arranged in chronological order to form the time-domain energy entropy sequence.

3. The method according to claim 2, characterized in that, The energy entropy value of each time-domain signal frame is determined by the following formula: in, The energy entropy value is... , Let i be the energy value of the i-th time-domain signal frame. ε represents the total number of frames in the valid signal phase, j represents the index of the valid signal phase, and ε represents the preset zero-prevention factor.

4. The method according to claim 1, characterized in that, The step of determining the identification result of the electromagnetic signal to be detected through sliding window correlation matching includes: A sliding window is set, the length of which is equal to the length of the target signal template; The sliding window is moved point by point on the global time-domain energy entropy sequence; For each sliding position, calculate the Pearson correlation coefficient between the sequence segment within the sliding window and the target signal template; When the Pearson correlation coefficient exceeds a preset detection threshold, it is determined that a signal matching the target signal template has been detected at the corresponding sliding position.

5. The method according to claim 4, characterized in that, The sliding window moves by a step of one energy entropy value each time.

6. The method according to claim 1, characterized in that, Before determining the identification result of the electromagnetic signal to be detected, the method further includes: Load multiple signal templates from the signal template library into the graphics processor's video memory; The global temporal energy entropy sequence is transferred to the graphics processor's video memory; The multi-threaded parallel computing architecture of the graphics processor is used to simultaneously calculate the sliding window correlation coefficient between the global time-domain energy entropy sequence and each of the signal templates.

7. The method according to claim 1, characterized in that, After determining the identification result of the electromagnetic signal to be detected, the method further includes: When a matching signal is detected, the peak value of the correlation coefficient when the match is successful is recorded; When the peak value of the correlation coefficient is in a preset high threshold range, the corresponding signal template in the signal template library is updated based on the time-domain energy entropy sequence within the current sliding window. When the peak value of the correlation coefficient is in the preset low threshold range and higher than the preset detection threshold, the time domain energy entropy sequence in the current sliding window is temporarily stored as a candidate new template, and when the cumulative occurrence of the candidate new template exceeds the preset number threshold, it is added to the signal template library as a new signal template.

8. The method according to claim 1, characterized in that, Before determining the target signal template from the signal template library, the method further includes: Each signal template in the signal template library is downsampled to construct a coarse-grained template index layer; The global temporal energy entropy sequence is downsampled by the same factor to obtain a coarse-grained detection sequence; The coarse-grained sequence to be detected is subjected to a sliding window correlation screening with the coarse-grained template index layer to select a set of candidate templates whose correlation coefficient exceeds the initial screening threshold. The sliding window correlation matching is performed only on signal templates in the candidate template set.

9. The method according to claim 1, characterized in that, The acquisition of time-domain data of the effective duration of a known type of electromagnetic signal includes: Effective signal detection is performed on the known types of electromagnetic signals to determine the start and end points; The time-domain data between the start and end points of the known type of electromagnetic signal is extracted and used as the time-domain data of the effective duration phase. The effective signal detection employs any one of the following methods: energy threshold method, double sliding window method, or short-time energy zero-crossing rate method.

10. An electromagnetic signal recognition system based on energy entropy template matching, characterized in that, The system includes: The signal acquisition module is used to acquire time-domain data of the effective duration of electromagnetic signals of known categories; The energy entropy sequence calculation module is used to generate a time-domain energy entropy sequence of the known category of electromagnetic signals based on the time-domain data, as a signal template; The template library storage module is used to store the signal templates and construct a signal template library; The global sequence acquisition module is used to acquire the global time-domain energy entropy sequence of the electromagnetic signal to be detected; The target template determination module is used to determine a target signal template from the signal template library; The decision output module is used to determine the identification result of the electromagnetic signal to be detected based on the global time-domain energy entropy sequence and the target signal template through sliding window correlation matching.

Images