Cyclically symmetric signal detection method combining multi-harmonic feature decision smoothing and medium

Abstract

The present application relates to a kind of cyclic stationary signal detection method and medium of combining multi-harmonic feature decision smoothing, it is related to signal processing technical field.The method includes: filtering based on the time series signal collected, and the cyclic spectrum of each time point is collected based on the preset acquisition frequency;For each cyclic spectrum, calculate weighted global probability score;For each cyclic spectrum, obtain its decision threshold, generate instantaneous decision result based on weighted global probability score and decision threshold;For the instantaneous decision result time series formed by instantaneous decision result, using sliding window smoothing processing, calculate the proportion of instantaneous decision result in sliding window that is true, judge whether to detect cyclic stationary signal.Compared with prior art, isolated false alarm caused by instantaneous interference can be effectively filtered out, so that a better balance between the reliability of feature and the stability of decision can be achieved, and the reliability of target detection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing, in particular to a cyclic stationary signal detection method combining multi-harmonic feature decision smoothing and a medium. BACKGROUND

[0002] When analyzing a cyclic stationary signal (such as helicopter noise, communication signal, etc.), the cyclic spectrum (or spectral correlation density) of the signal is usually calculated to reveal its hidden periodicity. After obtaining the cyclic spectrum, a reliable detector is needed to determine whether the target feature exists.

[0003] To improve the reliability of detection, the existing technology usually adopts a multi-harmonic feature fusion strategy. This strategy first identifies feature peaks at the fundamental and multiple harmonic frequencies, then fuses these discrete features into a single global probability score (P), and makes a decision based on this.

[0004] However, the above method cannot achieve a good balance between the reliability of the feature and the stability of the decision, resulting in limited performance and insufficient robustness in real and variable environments. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a cyclic stationary signal detection method combining multi-harmonic feature decision smoothing and a medium, which can achieve a better balance between the reliability of the feature and the stability of the decision.

[0006] In a first aspect, a cyclic stationary signal detection method combining multi-harmonic feature decision smoothing is provided in an embodiment, comprising:

[0007] Filtering based on the collected time series signal, and collecting the cyclic spectrum of each time point based on the preset collection frequency;

[0008] For each cyclic spectrum, calculating its weighted global probability score;

[0009] For each cyclic spectrum, obtaining its decision threshold, and based on the weighted global probability score and the decision threshold, generating an instantaneous decision result;

[0010] For the instantaneous decision result time series formed by the instantaneous decision result, using a sliding window smoothing process to calculate the proportion of the instantaneous decision result being true within the sliding window, and determining whether a cyclic stationary signal is detected.

[0011] In an embodiment, for any one cyclic spectrum, the method for calculating its weighted global probability score comprises:

[0012] Based on the adaptive threshold cyclic spectrum detection function, obtaining the spectral peak value of each expected cyclic spectrum cycle frequency band.

[0013] based on each spectrum peak value, calculating a normalized local probability score of the spectrum peak;

[0014] based on the reliability difference of each harmonic feature, assigning a corresponding weight to each local probability score representing different harmonic orders, and performing a weighted summation of all local probability scores based on the weights to calculate a weighted global probability score of the current cyclic spectrum; wherein the assigned weight monotonically decreases with the increase of the harmonic order.

[0015] In an embodiment, the cyclic spectrum detection function based on the adaptive threshold obtains spectrum peaks of a frequency band in which each of the plurality of expected cyclic spectrum cycle frequencies is located, including:

[0016]

[0017]

[0018] wherein, represents a spectrum peak value of a frequency band in which the i-th expected cyclic spectrum cycle frequency is located, 2≤i≤n, n represents the number of expected cyclic spectrum cycle frequencies; represents a cyclic spectrum detection function based on an adaptive threshold, used to obtain an amplitude of a cyclic spectrum correlation function along a frequency axis at each cycle frequency point; represents the k-th cycle frequency point in the frequency band in which the i-th expected cyclic spectrum cycle frequency is located, 1≤k≤K, K represents the number of cycle frequency points in the frequency band in which the i-th expected cyclic spectrum cycle frequency is located; B represents the frequency range of the entire cyclic spectrum; represents a frequency band set for the i-th expected cyclic spectrum cycle frequency, f represents a distribution frequency of the energy of the collected time series signal in the frequency spectrum, represents the resolution of f, represents a cyclic spectrum correlation function.

[0019] In an embodiment, based on each spectrum peak value, calculating a normalized local probability score of the spectrum peak includes: obtaining a dynamic threshold value of the current cyclic spectrum, and calculating the normalized local probability score of the spectrum peak based on the dynamic threshold value and the spectrum peak value; the dynamic threshold value is set based on a preset false alarm rate, and is the minimum value of the amplitudes in the maximum number of amplitudes of the preset false alarm rate in the current cyclic spectrum.

[0020] In an embodiment, obtaining a dynamic threshold value of the current cyclic spectrum, and calculating a normalized local probability score of the spectrum peak based on the dynamic threshold value and the spectrum peak value includes:

[0021]

[0022] wherein, represents the local probability score corresponding to the i-th expected cyclic spectral cycle frequency, represents the dynamic threshold value of the q-th cyclic spectrum, 1≤q≤Q, Q represents the number of cyclic spectra, represents the false alarm rate.

[0023] In an embodiment, the reliability difference based on the harmonic characteristics is that a corresponding weight is assigned to each local probability score representing different harmonic orders, and a weighted global probability score of the current cyclic spectrum is calculated by weighted sum of all local probability scores based on the weights, including:

[0024]

[0025] wherein, represents the weighted global probability score of the q-th cyclic spectrum; represents the weight of the local probability score, > > .

[0026] In an embodiment, for any cyclic spectrum, a decision threshold value is obtained, and based on the weighted global probability score and the decision threshold value, an instantaneous decision result is generated, including:

[0027] calculating the average amplitude of the cycle frequency points in the cyclic spectrum whose amplitudes reach the dynamic threshold value;

[0028] calculating the normalized decision threshold value based on the average amplitude and the corresponding dynamic threshold value;

[0029] generating an instantaneous decision result based on the weighted global probability score and the corresponding decision threshold value.

[0030] In an embodiment, the calculation of the average amplitude of the cycle frequency points in the cyclic spectrum whose amplitudes reach the dynamic threshold value includes:

[0031]

[0032] wherein, represents the average amplitude of the q-th cyclic spectrum, represents the amplitude of the u-th cycle frequency point in the q-th cyclic spectrum, 1≤u≤U, U represents the number of cycle frequency points in the q-th cyclic spectrum; represents 1 when the amplitude of the u-th cycle frequency point is greater than the corresponding dynamic threshold value, otherwise 0; represents the number of cycle frequency points in the q-th cyclic spectrum whose amplitudes are greater than the corresponding dynamic threshold value;

[0033] The calculation of the normalized decision threshold value based on the average amplitude and the corresponding dynamic threshold value includes:

[0034]

[0035] wherein, represents the decision threshold of the qth cyclic spectrum;

[0036] The generating of the instantaneous decision result based on the weighted global probability score and the corresponding decision threshold comprises:

[0037]

[0038] wherein, represents the instantaneous decision result of the qth cyclic spectrum.

[0039] In an embodiment, for the time sequence of the instantaneous decision result formed by the instantaneous decision result, a sliding window smoothing processing is adopted, and the number or proportion of the instantaneous decision result that is true in the sliding window is calculated to determine whether a cyclostationary signal is detected, comprising:

[0040] The number or proportion of the instantaneous decision result that is true in the sliding window is calculated, and if the number that is true reaches a preset number threshold or the proportion reaches a preset proportion threshold, it is determined that a cyclostationary signal is detected, wherein the width W of the sliding window satisfies W >= 4, and the step S satisfies 1 <= S <= W.

[0041] In a second aspect, an embodiment provides a computer readable storage medium, wherein the medium stores a program capable of being loaded and executed by a processor to perform the cyclostationary signal detection method of any one of the above embodiments.

[0042] The beneficial effects of the present application are:

[0043] Based on the cyclostationary signal detection method and medium provided in the present application, the sliding window smoothing processing is adopted, the prior knowledge that the real target signal must have "time continuity" is utilized, the time sequence composed of the generated instantaneous decision result is post-processed, the isolated false alarm caused by instantaneous interference can be effectively filtered out, so that the final false alarm rate is reduced to a very low level without sacrificing the sensitivity, and a better balance between the reliability of the feature and the stability of the decision can be achieved, and the reliability of target detection is improved. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a cyclostationary signal detection method flowchart of the present application in an embodiment;

[0045] Figure 2 is a method flowchart for calculating the weighted global probability score of any one cyclic spectrum.

[0046] Figure 3 is a schematic diagram of setting a frequency band based on a preset cycle frequency point according to an embodiment of the present application;

[0047] Figure 4 is a schematic diagram of a method of obtaining a decision threshold of any one cyclic spectrum and generating an instantaneous decision result based on a weighted global probability score and the decision threshold. DETAILED DESCRIPTION

[0048] The application will be described in further detail below with reference to the drawings. Like elements in different embodiments are denoted by like reference numerals. In the following embodiments, many details are described in order to provide a better understanding of the present application. However, one skilled in the art can easily recognize that some features can be omitted in different cases, or can be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification in order to avoid the core of the present application being overwhelmed by too much description, and it is not necessary to describe these related operations in detail for one skilled in the art based on the description in the specification and general technical knowledge in the art.

[0049] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various embodiments. Meanwhile, the steps or actions in the method description can also be sequentially adjusted or adjusted in a manner that is obvious to one skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment, and do not mean that the sequence is necessary, unless otherwise stated that a certain sequence must be followed.

[0050] The serial numbers of components in this document, such as "first", "second", etc., are only used to distinguish the described objects and do not have any sequence or technical meaning.

[0051] In order to facilitate the description of the inventive concept of the present application, the cyclic stationary signal analysis technology is briefly described below.

[0052] Current cyclic stationary signal analysis usually uses a multi-harmonic feature fusion strategy. This strategy first identifies feature peaks at the fundamental and multiple harmonic frequencies, then fuses these discrete features into a single global probability score (P), and makes a decision based on this.

[0053] However, the applicant found in the research that the above-mentioned cyclic stationary signal analysis has two deep technical defects, which leads to insufficient robustness in practical application, including:

[0054] First, the feature fusion method is blind. The existing fusion method generally adopts simple arithmetic average of each harmonic feature to obtain a decision threshold. However, this method is physically unreasonable. It ignores the basic physical priori that the harmonic energy of the target signal generally decays with the increase of the harmonic order. As a result, it gives the same importance to the low-order harmonic with strong energy, high signal-to-noise ratio and better reliability and the high-order harmonic with weak energy, easy to be contaminated by noise and poor reliability. This "equal treatment" fusion strategy leads to that the final global score P itself is not stable enough and is easily disturbed by random noise in the high frequency band, thereby reducing the robustness of the decision from the root.

[0055] Second, the decision result is easily affected by instantaneous interference and lacks a confirmation mechanism in the time dimension. The decision of the existing method completely depends on the global score calculated by a single data frame. This mechanism is very sensitive to isolated and short strong noise pulses. The signal features of a real and persistent target will certainly appear stably in continuous time. The false features caused by random noise are usually short and isolated. The existing method cannot effectively distinguish between the two cases. If a random noise happens to imitate the target feature at a certain time, the system will produce a false alarm. In order to suppress such false alarms, it is often necessary to increase the decision threshold, but this will also reduce the detection sensitivity to real weak signals, resulting in a missed report.

[0056] Based on the cyclic stationary signal detection method and medium provided in the present application, the sliding window smoothing processing is adopted, the priori knowledge that the real target signal must have "time continuity" is used, the time sequence composed of the instantaneous decision results is post-processed, the isolated false alarm caused by instantaneous interference can be effectively filtered out, the final false alarm rate can be reduced to a very low level without sacrificing the sensitivity, a better balance between the reliability of the feature and the stability of the decision can be achieved, and the reliability of the target detection is improved.

[0057] The present application provides a cyclic stationary signal detection method combined with multi-harmonic feature decision smoothing. Please refer to Figure 1 , which comprises:

[0058] Step S10, filtering based on the collected time sequence signal, and collecting the cyclic spectrum of each time point based on the preset acquisition frequency.

[0059] The collected time sequence signal can be a helicopter noise, a communication signal, etc. The cyclic spectrum of each time point is collected based on the preset acquisition frequency after filtering the signal, so as to obtain Q cyclic spectrums.

[0060] Step S20, for each cyclic spectrum, calculate its weighted global probability score.

[0061] In one embodiment, please refer to Figure 2 For any one cyclic spectrum, the method for calculating its weighted global probability score can include:

[0062] Step S201, based on the adaptive threshold cyclic spectrum detection function, obtain the spectral peak value of the frequency band where each of the plurality of expected cyclic spectrum cycle frequencies is located.

[0063] In the embodiments of the present application, the fixed threshold is abandoned, and based on the adaptive threshold (the 90% quantile of each data point is used as the threshold) cyclic spectrum detection function, the spectral peak value of the frequency band where each of the plurality of expected cyclic spectrum cycle frequencies is located is obtained, so that the threshold can automatically adapt to the fluctuation of the current background noise, and has strong adaptability.

[0064] Please refer to Figure 3 , the expected cycle frequencies are multiplied, in the diagram, the first expected cycle frequency is , the frequency band where it is located is B1; the second expected cycle frequency =2 , the frequency band where it is located is B2; the second expected cycle frequency =3 , the frequency band where it is located is B3.

[0065] In one embodiment, step S201 can be represented as:

[0066]

[0067]

[0068] Wherein, represents the spectral peak value of the frequency band where the ith expected cyclic spectrum cycle frequency is located, 2≤i≤n, n represents the number of expected cyclic spectrum cycle frequencies; represents the adaptive threshold cyclic spectrum detection function, which is used to obtain the amplitude of the cyclic spectrum correlation function along the frequency axis for each cycle frequency point; represents the kth cycle frequency point in the frequency band where the ith expected cyclic spectrum cycle frequency is located, 1≤k≤K, K represents the number of cycle frequency points in the frequency band where the ith expected cyclic spectrum cycle frequency is located; B represents the spectral frequency range of the entire cyclic spectrum; represents the frequency band set for the ith expected cyclic spectrum cycle frequency, f represents the distribution frequency of the energy of the collected time series signal on the frequency spectrum, represents the resolution of f, represents the cyclic spectrum correlation function.

[0069] Step S202, based on each spectrum peak peak value, calculate the normalized local probability score of the spectrum peak.

[0070] In an embodiment, in order to reflect the significant degree that the spectrum peak peak value exceeds the noise floor, step S202 includes: obtaining a dynamic threshold value of the current cycle spectrum, and calculating the normalized local probability score of the spectrum peak based on the dynamic threshold value and the spectrum peak peak value. Wherein, the dynamic threshold value is set based on a preset false alarm rate, and is the minimum value of the maximum amplitude number of the preset false alarm rate in the current cycle spectrum. For example, if there are 100 cycle spectrum cycle frequency points in the current cycle spectrum, the amplitude of the 100 cycle spectrum cycle frequency points is obtained. If the preset false alarm rate is 10%, the maximum amplitude of 100*10%=10 is obtained, and the minimum amplitude of the 10 maximum amplitudes is taken as the dynamic threshold value of the current cycle spectrum.

[0071] Obtaining the dynamic threshold value of the current cycle spectrum, and calculating the normalized local probability score of the spectrum peak based on the dynamic threshold value and the spectrum peak peak value, includes:

[0072]

[0073] Wherein, represents the local probability score corresponding to the ith expected cycle spectrum cycle frequency, represents the dynamic threshold value of the qth cycle spectrum, 1≤q≤Q, Q represents the number of cycle spectra, represents the false alarm rate.

[0074] The local probability score reflects the significant degree that the spectrum peak peak value exceeds the dynamic threshold value corresponding to it , that is, it reflects the significant degree that the spectrum peak peak value exceeds the noise floor.

[0075] Step S203, based on the reliability difference of each harmonic feature, assign a corresponding weight to each local probability score representing different harmonic orders, and perform weighted summation on all local probability scores based on the weight to calculate the weighted global probability score of the current cycle spectrum. Wherein, the assigned weight monotonically decreases with the increase of the harmonic order.

[0076] In an embodiment, step S203 can be represented as:

[0077]

[0078] Wherein, represents the weighted global probability score of the qth cycle spectrum; represents the weight of the local probability score, >​ .

[0079] As can be appreciated by those skilled in the art, from P1 to P n , the harmonic order is increasing and the energy is decreasing, therefore, the decreasing weight-based allocation makes the fundamental and low-order harmonics be given higher weights, which is more consistent with the physical priori that the harmonic energy of a target such as a helicopter generally decays with the increase of the harmonic order, so that the detection result is more accurate. Thus, the weighted global probability score of the embodiments of the present application synthesizes the evidence from multiple harmonics, which is much more reliable than a single feature, and when a certain harmonic is weak, other stronger harmonics can still support the final decision, so that a more accurate detection result is obtained.

[0080] In an embodiment, an attenuation model is used to determine the weight of the local probability score.

[0081] In step S30, the decision threshold of each cyclic spectrum is obtained, and the instantaneous decision result is generated based on the weighted global probability score and the decision threshold.

[0082] In an embodiment, please refer to Figure 4 , for any cyclic spectrum, the decision threshold is obtained, and the instantaneous decision result is generated based on the weighted global probability score and the decision threshold, including:

[0083] In step S301, the average amplitude of the cyclic frequency points with amplitudes reaching the dynamic threshold value in the cyclic spectrum is calculated.

[0084] In an embodiment, step S301 can be expressed as:

[0085]

[0086] wherein, represents the average amplitude of the qth cyclic spectrum, represents the amplitude of the u th cyclic frequency point in the q th cyclic spectrum, 1≤u≤U, U represents the number of cyclic frequency points in the q th cyclic spectrum; represents 1 when the amplitude of the u th cyclic frequency point is greater than the corresponding dynamic threshold value, otherwise 0; represents the number of cyclic frequency points in the q th cyclic spectrum whose amplitudes are greater than the corresponding dynamic threshold value.

[0087] In step S302, the normalized decision threshold is calculated based on the average amplitude and the corresponding dynamic threshold value.

[0088] In an embodiment, step S302 can be expressed as:

[0089] ​

[0090] wherein, denotes the decision threshold of the qth cyclic spectrum.

[0091] The decision threshold reflects the overall intensity level of the effective signal components in the current signal.

[0092] Step S303, based on the weighted global probability score and the corresponding decision threshold, generates an instantaneous decision result.

[0093] The weighted global probability score and the corresponding decision threshold are compared, thereby generating the instantaneous decision result of the current data frame (cyclic spectrum). In one embodiment, it can be denoted as:

[0094]

[0095] wherein, denotes the instantaneous decision result of the qth cyclic spectrum.

[0096] In the above decision result, 1 indicates that true is detected, and 0 indicates that false is not detected.

[0097] Step S40, for the instantaneous decision result time sequence formed by the instantaneous decision result, a sliding window smoothing processing is adopted to calculate the proportion of the instantaneous decision result being true within the sliding window, and it is judged whether the cyclostationary signal is detected.

[0098] In one embodiment, a post-processing filter of a decision level is introduced, which does not operate on the signal itself, but processes the instantaneous decision result time sequence composed of continuous multiple data frames (cyclic spectra) , specifically including: calculating the number or proportion of the instantaneous decision result being true within the sliding window, if the number of being true reaches a preset number threshold or the proportion reaches a preset proportion threshold, it is judged that the cyclostationary signal is detected, the width W of the sliding window satisfies W≥4, and the step S satisfies 1≤S≤W.

[0099] For example, in the case of the width W of the sliding window being 5, if the preset number threshold is 2, then the number of the instantaneous decision result being true within the sliding window reaching 2 indicates that the cyclostationary signal is detected; if the preset number threshold is 3, then the number of the instantaneous decision result being true within the sliding window reaching 3 indicates that the cyclostationary signal is detected; if the preset proportion threshold is 50%, then the proportion of the number of the instantaneous decision result being true within the sliding window reaching 50% (i.e. 3 or more) indicates that the cyclostationary signal is detected.

[0100] Due to a real, persistent target, a series of instantaneous decision "1" will inevitably be generated in continuous time; while false alarm caused by a random, short-term noise interference is difficult to form a persistent response in time. Therefore, this step takes advantage of the "time continuity" prior knowledge of the target, and can effectively filter out false alarms caused by instantaneous interference, thereby reducing the final false alarm rate to a very low level without sacrificing sensitivity, so that a better balance between feature reliability and decision stability can be achieved, and the reliability of target detection is improved.

[0101] In the embodiments of the present application, in the first aspect, due to the use of sliding window smoothing processing, the prior knowledge that a real target signal must have "time continuity" is utilized, and the time sequence composed of the generated instantaneous decision results is post-processed, which can effectively filter out isolated false alarms caused by instantaneous interference, thereby reducing the final false alarm rate to a very low level without sacrificing sensitivity, so that a better balance between feature reliability and decision stability can be achieved, and the reliability of target detection is improved.

[0102] In the second aspect, since the introduced global weighted probability score calculation is based on the physical prior that "harmonic energy decays with the number of times", different weights are given to the local probability scores of different harmonics, which is essentially different from simple arithmetic mean. Therefore, this method can intelligently focus on more reliable low-order harmonics and suppress high-order harmonic noise interference, so that the instantaneous decision result is more robust. This directly improves the detection rate under low signal-to-noise ratio (such as far-field helicopter noise), and effectively avoids false negatives caused by weak partial harmonic features.

[0103] In an embodiment of the present application, a computer readable storage medium is provided, and a program is stored on the storage medium. The stored program includes a method capable of being loaded and processed by a processor in any of the above embodiments.

[0104] Those skilled in the art can understand that all or part of the functions of the various methods in the above embodiments can be realized by hardware or by a computer program. When all or part of the functions in the above embodiments are realized by a computer program, the program can be stored in a computer readable storage medium, which can include read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc. The above functions are realized by executing the program by a computer. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, the above functions are realized. In addition, when all or part of the functions in the above embodiments are realized by a computer program, the program can also be stored in a server, another computer, a disk, an optical disk, a flash disk or a mobile hard disk, etc. The program is downloaded or copied and saved to the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, the above functions are realized.

[0105] The above application of specific examples to illustrate the present invention, is only used to help understand the present invention, and does not limit the present invention. For the skilled in the art to which the present invention belongs, according to the idea of the present invention, several simple deductions, deformation or replacement can be made.

Claims

1. A method for detecting cyclostationary signals by combining multi-harmonic characteristic decision smoothing, characterized in that, include: The collected time-series signal is filtered, and the cyclic spectrum of each time point is collected based on the preset acquisition frequency. The collected time-series signal is helicopter noise or communication signal. For each cyclic spectrum, calculate its weighted global probability score; For each cyclic spectrum, obtain its decision threshold, and generate an instantaneous decision result based on the weighted global probability score and the decision threshold; For the time series of instantaneous decision results formed by instantaneous decision results, a sliding window smoothing process is used to calculate the proportion of instantaneous decision results that are true within the sliding window and to determine whether a cyclic stationary signal is detected. For any cyclic spectrum, methods for calculating its weighted global probability score include: Based on an adaptive threshold cyclic spectrum detection function, the peak-to-peak value of the spectral frequency band of each of the multiple expected cyclic spectrum cyclic frequencies is obtained. Based on the peak value of each spectral peak, calculate the normalized local probability score of that spectral peak; Based on the reliability differences of each harmonic characteristic, a corresponding weight is assigned to each local probability score representing different harmonic orders, and all local probability scores are weighted and summed based on the weights to calculate the weighted global probability score of the current cyclic spectrum; wherein, the assigned weights decrease monotonically with the increase of the harmonic order. The method of calculating the normalized local probability score of each spectral peak based on its peak value includes: obtaining the dynamic threshold value of the current cyclic spectrum, and calculating the normalized local probability score of the spectral peak based on the dynamic threshold value and the peak value; the dynamic threshold value is set based on a preset false alarm rate and is the minimum value among the maximum amplitude values ​​of the preset false alarm rate in the current cyclic spectrum.

2. The method for detecting cyclic stationary signals as described in claim 1, characterized in that, The adaptive threshold-based cyclic spectrum detection function obtains the spectral peaks of the frequency bands where the cyclic frequencies of multiple expected cyclic spectra are located, including: in, Let represent the peak-to-peak value of the frequency band in which the i-th expected cyclic spectrum cycle frequency is located, 2≤i≤n, where n represents the number of expected cyclic spectrum cycle frequencies; This represents a cyclic spectrum detection function based on an adaptive threshold, used to obtain the amplitude of the integral of the cyclic spectrum correlation function along the frequency axis at each cyclic frequency point; B represents the k-th cyclic frequency point in the frequency band where the cyclic frequency of the i-th expected cyclic spectrum is located, 1≤k≤K, where K represents the number of cyclic frequency points in the frequency band where the cyclic frequency of the i-th expected cyclic spectrum is located; B represents the frequency range of the entire cyclic spectrum. Let f represent the frequency band set for the i-th expected cyclic spectrum cycle frequency, and let f represent the frequency distribution of the energy of the acquired time-series signal in the spectrum. This represents the resolution of f. This represents the cyclic spectrum correlation function.

3. The method for detecting cyclic stationary signals as described in claim 2, characterized in that, The process of obtaining the dynamic threshold value of the current cyclic spectrum and calculating the normalized local probability fraction of the spectral peak based on the dynamic threshold value and the peak value includes: in, Let represent the local probability fraction corresponding to the cycle frequency of the i-th expected cyclic spectrum. This represents the dynamic threshold value of the q-th cyclic spectrum, where 1 ≤ q ≤ Q, and Q represents the number of cyclic spectra. This indicates the false alarm rate.

4. The method for detecting cyclic stationary signals as described in claim 3, characterized in that, The aforementioned method, based on the reliability differences of each harmonic characteristic, assigns a corresponding weight to each local probability score representing different harmonic orders, and performs a weighted summation of all local probability scores based on these weights to calculate the weighted global probability score of the current cyclic spectrum, including: in, This represents the weighted global probability score of the q-th cyclic spectrum; The weights represent the local probability scores. > > .

5. The method for detecting cyclic stationary signals as described in claim 4, characterized in that, For any cyclic spectrum, obtain its decision threshold. Based on the weighted global probability score and the decision threshold, generate an instantaneous decision result, including: Calculate the average amplitude of the cyclic frequency points in the cyclic spectrum where the amplitude reaches the dynamic threshold; Calculate the normalized decision threshold based on the average amplitude and the corresponding dynamic threshold; Based on the weighted global probability score and the corresponding decision threshold, an instantaneous decision result is generated.

6. The method for detecting cyclic stationary signals as described in claim 5, characterized in that, The calculation of the average amplitude of the cyclic frequency points in the cyclic spectrum where the amplitude reaches the dynamic threshold includes: in, This represents the average amplitude of the q-th cyclic spectrum. Let U represent the amplitude of the u-th cyclic frequency point in the q-th cyclic spectrum, where 1 ≤ u ≤ U, and U represents the number of cyclic frequency points in the q-th cyclic spectrum. The value is 1 if the amplitude at the u-th cycle frequency point is greater than the corresponding dynamic threshold value, and 0 otherwise. This represents the number of times the amplitude of the cyclic frequency point in the q-th cyclic spectrum is greater than the corresponding dynamic threshold value; The calculation of the normalized decision threshold based on the average amplitude and the corresponding dynamic threshold includes: in, This represents the decision threshold for the q-th cyclic spectrum; The generation of instantaneous decision results based on the weighted global probability score and the corresponding decision threshold includes: in, This represents the instantaneous decision result of the q-th cyclic spectrum.

7. The method for detecting cyclic stationary signals as described in claim 5, characterized in that, For the time series of instantaneous decision results formed by instantaneous decision results, a sliding window smoothing process is used to calculate the number or proportion of true instantaneous decision results within the sliding window, and to determine whether a cyclic stationary signal is detected, including: Calculate the number or percentage of true results within the sliding window. If the number of true results reaches a preset quantity threshold or the percentage reaches a preset proportion threshold, then a cyclic steady-state signal is detected. The width W of the sliding window satisfies W≥4, and the step size S satisfies 1≤S≤W.

8. A computer-readable storage medium, characterized in that, The medium stores a program that can be loaded by a processor and executed as the cyclic stationary signal detection method as described in any one of claims 1 to 7.

Images