Support vector machine interference detection method based on feature selection and maximum posteriori

Abstract

The invention provides a support vector machine interference detection method based on feature selection and maximum posteriori, multi-dimensional interference features are weighted and optimized by introducing CUSUM statistics, efficient screening and enhancement of key features are realized, a maximum posteriori probability criterion is innovatively introduced for adaptively optimizing a classification threshold value of a support vector machine, and the accuracy of the support vector machine interference detection is improved. The threshold value is dynamically adjusted to adapt to a complex and changeable electromagnetic environment, so that the accuracy and the stability of interference detection are improved. According to the method, the feature discrimination capability and the adaptive response of a detection system are remarkably enhanced, the limitation of traditional fixed threshold and single feature selection is broken through, efficient, accurate and robust interference detection is realized, and the method has relatively high innovativeness and practical application value.

Description

Technical Field

[0001] This invention relates to the field of communication signal processing and interference detection, specifically to a support vector machine interference detection method based on feature selection and maximum a posteriori. Background Technology

[0002] In Heterogeneous Self-Organizing Networks (H-SONs), multiple access points, including macro base stations, micro base stations, pico base stations, and relay nodes, operate collaboratively to construct a densely distributed, highly dynamic, and adaptive wireless access architecture. H-SONs, as a crucial technological foundation for achieving high spectrum utilization, flexible networking, and intelligent resource scheduling in next-generation communication systems, have been widely applied in 5G, future tactical communications, low-altitude intelligent networks, and large-scale industrial IoT. However, due to the high deployment density of heterogeneous nodes, irregular topology, and frequent dynamic changes in user access, the system faces challenges such as intensified competition for air interface resources, susceptibility to electromagnetic interference, and severe spectrum congestion. It is particularly vulnerable to threats from various intentional and unintentional interference sources in open environments.

[0003] Interference signals not only disrupt the communication performance of user terminals but can also affect the stability of critical control links, leading to serious consequences such as increased communication reconfiguration overhead and frequent connection interruptions. Especially in applications with extremely high requirements for communication reliability, such as military, emergency response, and industrial control, the ability to detect electromagnetic interference quickly, accurately, and with low false alarm rates has become one of the core requirements for the operation of heterogeneous self-organizing networks.

[0004] Current mainstream interference detection methods, such as energy detection, peak spectrum analysis, or statistical detection algorithms, have been used for interference monitoring in some heterogeneous network environments. However, these methods generally suffer from problems such as fixed detection thresholds, insensitivity to low signal-to-interference ratios (SINR), limited feature dimensions, and poor generalization ability. In complex dynamic networks, rapid changes in channel conditions lead to enhanced feature fluctuations, making fixed-threshold strategies prone to misjudgment. Especially in heterogeneous self-organizing networks, the distribution of channel states among nodes is extremely uneven, and interference characteristics differ significantly under different geographical, frequency band, and load conditions. Static models and fixed decision strategies are insufficient to meet the dynamic detection requirements in complex environments. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a support vector machine interference detection method based on feature selection and maximum a posteriori (MAP). By introducing the CUSUM statistic to weight and optimize multi-dimensional interference features, it achieves efficient screening and enhancement of key features. Furthermore, it innovatively introduces the maximum a posteriori (MAP) criterion for adaptive optimization of the SVM classification threshold, enabling dynamic threshold adjustment to adapt to complex and changing electromagnetic environments, thereby improving the accuracy and stability of interference detection. The technical solution provided by this invention is as follows:

[0006] A support vector machine-based interference detection method based on feature selection and maximum a posteriori includes the following steps:

[0007] Step 1: Continuously sample the wireless signal and divide the received discrete time-domain signal into frames of length N;

[0008] Step 2: For each frame of signal, extract features from multiple dimensions including time domain, frequency domain, time-frequency domain, and statistical domain;

[0009] Step 3: Feature selection is performed using an optimization strategy that integrates random forest importance evaluation with prior mutation detection correlation weighting.

[0010] Step 4: Train the classifier;

[0011] Step 5: Introduce an adaptive decision threshold adjustment mechanism based on maximum a posteriori probability to distinguish between interference and non-interference.

[0012] Preferably, in the time domain, mean square power features, peak-to-average power ratio features, and kurtosis are extracted; in the frequency domain, peak power spectral density, maximum spectral peak, spectral entropy, and 3dB bandwidth are extracted; in the time-frequency domain, maximum time-frequency energy and time-frequency spectral entropy are extracted; and in the statistical domain, fractal dimension is calculated.

[0013] Preferably, feature selection is based on random forest importance scores. Stability score Redundancy penalty factor and cumulative correlation score with CUSUM The obtained weighted importance score The following was carried out:

[0014]

[0015] in , , Measuring the global discriminative power of features Reflects temporal stability, Adjusting the sensitivity to mutation interference, Control redundancy penalty; sort features by weighted importance score from highest to lowest, and select the highest-scoring feature subset S such that:

[0016]

[0017] in .

[0018] Preferably, the specific process of step 4 is as follows:

[0019] Let the training sample set after feature selection optimization be:

[0020]

[0021] in This represents the d-dimensional optimal feature vector of the i-th training sample. These represent the category labels "no interference" and "with interference" respectively, and M is the total number of training sample signal frames;

[0022] The Gaussian radial basis function kernel is selected as the kernel mapping for the support vector machine:

[0023]

[0024] in The optimal value for the kernel width parameter is determined through grid search combined with cross-validation.

[0025] The optimization objective of support vector machines is: ;

[0026] The constraints are:

[0027] Where w is the weight vector and b is the bias term. For kernel mapping, As a penalty factor, These are slack variables;

[0028] The sequential minimum optimization algorithm is used to solve this convex quadratic programming problem, yielding the optimal set of support vectors and the corresponding Lagrange multipliers. The final decision function obtained after training:

[0029]

[0030] The decision function outputs continuous real numbers, with the positive or negative sign indicating the category and the absolute value representing the classification confidence.

[0031] Preferably, in step 4, during the training phase, the probability density estimates of the support vector machine output under the interference class and the non-interference class are additionally saved. and .

[0032] Preferably, the specific process of step 5 is as follows:

[0033] The feature vector after feature selection is: The decision function obtained from training the support vector machine is: The probability density functions for the interference class and the non-interference class are respectively and The prior probabilities of the samples are respectively and The maximum a posteriori probability decision formula is:

[0034]

[0035] This judgment is equivalently transformed into a comparison. With the optimal threshold Size;

[0036] The fitted probability density functions of the interference and non-interference classes are obtained by using kernel density estimation or Gaussian mixture models. and Calculate at the threshold The following are the probabilities of missed detection and false alarms:

[0037]

[0038] Construction optimization problem: ;

[0039] in The coefficient is used to adjust the importance of missed detections and false alarms in the optimization objective; by solving this optimization problem, the following is obtained: It will adaptively change according to the scoring distribution characteristics of different interference scenarios;

[0040] The final judgment formula is:

[0041]

[0042] Complete the distinction between interference and non-interference states.

[0043] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0044] This invention introduces the CUSUM statistic in the feature selection stage to weight the importance of features based on correlation, effectively enhancing the sensitivity to features affected by mutations. In the classification stage, a support vector machine is used to perform nonlinear mapping on multi-domain features, achieving high-precision classification and decision-making for multiple types and forms of interference. Compared with traditional single-domain or fixed threshold methods, it has a higher detection probability and a lower false alarm rate.

[0045] This invention introduces Bayesian inference into the machine learning decision stage by applying MAP-based post-processing to the SVM output score, enabling the classification threshold to be adaptively adjusted in different interference environments. This significantly improves the detection probability and reduces the false alarm probability, making it particularly suitable for highly dynamic wireless communication environments with multiple interference types, such as heterogeneous self-organizing networks.

[0046] The overall algorithm of this invention consists of feature extraction, CUSUM-based feature selection optimization, SVM classification decision, and MAP threshold adaptive adjustment. Its computational complexity is lower than that of deep network methods and it is easy to implement in parallel. It can run in real time on heterogeneous self-organizing network nodes, meet the needs of dynamic interference detection and rapid response, and has good engineering feasibility and promotion value. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0048] Figure 1 This is a system scenario diagram of the present invention;

[0049] Figure 2 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Example 1: A support vector machine-based interference detection method based on feature selection and maximum a posteriori (MAP). This method is applicable to, for example... Figure 1 In the system scenario shown, the algorithm flow is as follows: Figure 2 As shown, it includes the following steps:

[0053] Step 1: Preprocess the received signal. The receiver continuously samples the wireless signal and divides the received discrete-time signal into frames of length N. Let the sampled signal be represented as follows: Assume the discrete-time signal collected by the wireless receiver is... ,in For useful signal samples, For interference signal samples, It is additive white Gaussian noise.

[0054] Step 2: After preprocessing, the received discrete signal is divided into non-overlapping frames of length N. Each frame serves as an independent analysis unit for feature extraction, feature selection, and classification. For each frame, features are extracted from multiple dimensions, including time domain, frequency domain, time-frequency domain, statistics, and complexity, to ensure coverage of diverse manifestations of different types of interference, forming a complete feature vector set as the basis for subsequent feature selection and support vector machine classifier input.

[0055] In the time domain, the extracted mean square power feature is represented as: This feature can reflect the power increase phenomenon caused by strong interference signals or pulse interference.

[0056] Extract the ratio of the maximum amplitude to the average power within the frame to construct the peak-to-average power ratio feature: It reflects the short-term burst intensity of the signal.

[0057] Extract kurtosis to measure the sharpness of the signal amplitude distribution: This feature provides high sensitivity to sudden interference.

[0058] In the frequency domain, each frame of signal is subjected to a Fast Fourier Transform to obtain the frequency domain spectrum. Extract the peak power spectral density to measure the energy intensity of the main frequency components: .

[0059] Calculate the normalized characteristic of the maximum spectral peak to characterize the sharpness of the spectrum: .

[0060] Extracting spectral entropy: ,in To normalize the power spectrum distribution, reflecting the uniformity of the spectral energy distribution, m is the total number of spectral components used to calculate the spectral entropy.

[0061] The 3dB bandwidth is estimated using the energy envelope method: ,satisfy ,in This is the upper boundary frequency (where the power drops to half of its peak value), meaning it is first reached when the power spectrum decreases from its peak value towards higher frequencies. ; This is the lower boundary frequency (where the power drops to half of its peak value), meaning it first reaches this frequency when the power spectrum decreases from its peak value towards lower frequencies. .

[0062] In the time-frequency domain, the time-frequency matrix is ​​obtained using the short-time Fourier transform. Extracting the maximum time-frequency energy is used to capture strong interference concentrated in local time frequencies. .

[0063] Extraction time spectral entropy: ,in This is a normalized time-frequency energy distribution used to describe the discreteness and complexity of the time-frequency distribution.

[0064] In terms of statistics and complexity characteristics, fractal dimension is calculated to characterize the complexity of a signal's temporal structure. A higher fractal dimension indicates more non-stationarity and richer detail variations in the signal's morphology. The signal is normalized and mapped to a two-dimensional unit space, and its curve length is used to estimate the fractal dimension. Given a discrete signal sequence... Normalize the timeline: ;

[0065] Standardize the amplitude to [0,1]: ;

[0066] The path length of the signal point in the two-dimensional plane is: ;

[0067] because ,have to ;

[0068] According to Sevcik's fractal dimension formula Where L is the standardized path length. The formula for calculating the final fractal dimension, which is the maximum possible path length on the normalized plane, is:

[0069]

[0070] In summary, each frame constitutes a complete multi-domain feature vector: .

[0071] Each feature is normalized and standardized to adapt to the input of the subsequent feature selector and classifier. This multi-domain fusion feature construction method can fully extract significant features of interference in the signal, such as energy features, spectral features, statistical changes, and nonlinear changes, and provide complete high-dimensional information support for subsequent feature selection.

[0072] Step 3: The feature selection module employs an optimization strategy that integrates random forest importance evaluation with prior mutation detection correlation weighting to ensure that the selected features not only possess global discriminative ability in heterogeneous self-organizing network interference environments but also maintain high sensitivity to sudden interference. First, a candidate feature set is constructed from the frame signal processed by the feature extraction module. Let the sampled signal of the i-th frame be... ,in Indicates the frame index. The candidate feature set obtained after extraction is the intra-frame sampling point index. , where f represents the feature index and D is the total number of features. Definition This is the category label for this frame. During the training phase, a random forest model containing T trees is used to calculate the reduction in impurity for each feature across all trees, and this reduction is combined with the number of samples per node to form the Mean Impurity Reduction (MDI) metric, expressed as:

[0073]

[0074] in Let f represent the set of split nodes in the t-th tree that use feature f. This represents the reduction in node impurity. Let be the number of node samples, and M be the total number of samples. The baseline importance score is obtained by averaging this metric across all trees.

[0075]

[0076] The dimensionless global discriminant score was obtained after normalization. .

[0077] To enhance the robustness of features across time, this invention introduces a time stability metric. The training data is divided into B consecutive blocks, and the mean value of the features is calculated within each block. Then calculate the global mean. and inter-block variance Therefore, the stability score is defined as follows:

[0078]

[0079] in As a stability adjustment factor, it is usually set to 10. -8 , The adjustment coefficient controls the sensitivity of stability to the score. When a feature exhibits minimal fluctuation across blocks, the stability score approaches 1, thus prioritizing the retention of these highly stable features during feature selection.

[0080] To mitigate the adverse effects of feature redundancy, this invention calculates the maximum absolute correlation coefficient between features and incorporates it as a redundancy penalty term into the scoring system.

[0081]

[0082] To significantly improve sensitivity to mutation-related interference, this invention introduces correlation guidance based on CUSUM mutation statistics for the first time in the feature selection stage:

[0083] First, calculate the mean of the signal in this frame. As a reference mean under interference-free conditions, and defining the deviation at each sampling point. Based on this, a one-sided CUSUM statistic series is constructed:

[0084]

[0085] in This is the drift term, used to control the sensitivity of the statistic; a smaller value makes it more sensitive to abrupt changes. This statistic accumulates rapidly and exceeds a preset detection threshold when the signal mean changes significantly, thus reflecting the strength of abrupt interference in the frame. To quantify the CUSUM result into a scalar that can be associated with the feature set, this invention calculates the maximum statistic value for the frame:

[0086]

[0087] This value can be considered as the amplitude of the abrupt change response in the i-th frame, effectively distinguishing between stable background noise and transient interference frames. Subsequently, the candidate feature set... For each feature f in the sequence, calculate its relationship with the CUSUM response sequence. Pearson correlation coefficient:

[0088]

[0089] in and These are the mean values ​​of the feature and the CUSUM response, respectively. The correlation coefficient reflects the degree of linear association between the feature and sensitivity to mutational interference. A value close to 1 indicates that the feature exhibits a significant response when mutational disturbances occur; when... When the value is close to 0, it indicates that the feature is not sensitive to mutation interference.

[0090] Finally, the importance score of the random forest is based on comprehensive features. Stability score With redundancy penalty factor Based on this, the weighted importance score is defined by introducing CUSUM relevance weighting:

[0091]

[0092] in , , Measuring the global discriminative power of features Reflects temporal stability, Adjusting the sensitivity to mutation interference, Control redundancy penalty. When there is no CUSUM bootloader... Finally, based on the weighted importance scores, the features are sorted from highest to lowest, and the highest-scoring feature subset S is selected such that:

[0093]

[0094] in Using this subset as input to the subsequent classification module, adaptive optimization of feature selection was achieved by innovatively combining the prior correlation of CUSUM mutation detection with the importance assessment and stability analysis of random forest in the feature selection stage.

[0095] Step 4: After completing the feature extraction and feature selection modules, this invention enters the classifier training phase. Let the training sample set after feature selection optimization be:

[0096]

[0097] in This represents the d-dimensional optimal feature vector of the i-th training sample. These represent the "no interference" and "interference" category labels, respectively, and M is the total number of training sample signal frames used for initial feature calculation. To address the highly nonlinear distribution of interference features, this invention selects the Gaussian radial basis kernel function as the kernel mapping for the support vector machine:

[0098]

[0099] in The kernel width parameter is determined using a combination of grid search and cross-validation. The optimization objective of Support Vector Machine (SVM) is: ;

[0100] The constraints are:

[0101] Where w is the weight vector and b is the bias term. For kernel mapping, As a penalty factor, Let these be slack variables. The Sequential Minimum Optimization (SMO) algorithm is used to solve this convex quadratic programming problem, yielding the optimal set of support vectors and their corresponding Lagrange multipliers. The final decision function obtained through training is:

[0102]

[0103] The decision function outputs continuous real numbers, where the sign indicates the class assignment and the absolute value represents the classification confidence. This invention additionally stores the probability density estimates of the SVM output in both the interference and non-interference classes during the training phase. and This provides necessary prior data support for the next step of adaptive threshold adjustment based on maximum posterior probability.

[0104] Step 5: To improve the classifier's adaptability under different signal-to-noise ratios and interference patterns, this invention introduces an adaptive decision threshold adjustment mechanism based on maximum a posteriori probability to overcome the shortcomings of traditional support vector machines, such as fixed decision boundaries and sensitivity to changes in data distribution. Let the feature vector after feature selection be... The decision function obtained from training the support vector machine is: .

[0105] In traditional methods, when Determined as a interference hypothesis Conversely, the non-interference hypothesis. However, because the statistical characteristics of interference signals change dynamically with variations in signal-to-noise ratio, modulation method, and propagation environment, a fixed threshold can lead to a lower probability of missed detection in certain situations. or false alarm probability The performance of the continuous scores output by the SVM increases significantly, making it impossible to guarantee optimal decision performance. Therefore, this invention introduces the Maximum A posteriori (MAP) decision principle, which transforms the continuous scores from the SVM output into a single decision. This is transformed into a statistical decision metric. Let the probability density functions of the interference class and the non-interference class be respectively... and The prior probabilities of the samples are respectively and The MAP decision criterion is:

[0106]

[0107] This judgment formula can be equivalently transformed into a comparison. With an optimal threshold The size of. To solve for... This invention first utilizes a validation set to... Modeling the distributions for the two classes of samples can be done using kernel density estimation (KDE) or Gaussian mixture model (GMM) fitting. and And calculate at the threshold The following are the probabilities of missed detection and false alarms:

[0108]

[0109] Then construct the optimization problem.

[0110]

[0111] in This is a weighting factor used to adjust the importance of missed detections and false alarms in the optimization objective. When The tendency is to reduce the false alarm rate when The time bias is to reduce the false negative rate. By solving this optimization problem, the following is obtained: It will adaptively change according to the score distribution characteristics of different interference scenarios. The final SVM decision formula becomes:

[0112]

[0113] The system distinguishes between interfered and non-interference states and outputs the results to the interference management module in real time. This invention designs an interference detection scheme that can operate in real time in heterogeneous self-organizing networks, possessing the advantages of adaptive decision boundaries and stable detection performance.

[0114] Example 2: The computer-readable storage medium of this example stores a computer program that, when executed by a processor, implements the steps of the support vector machine interference detection method based on feature selection and maximum a posteriori in Example 1.

[0115] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.

[0116] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0117] Example 2: The computer device of this example includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the support vector machine interference detection method based on feature selection and maximum a posteriori of Example 1.

[0118] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0119] Those skilled in the art will clearly understand that each implementation can be achieved using software plus the necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0120] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A support vector machine-based interference detection method based on feature selection and maximum a posteriori (MAP), characterized in that, Includes the following steps: Step 1: Continuously sample the wireless signal and divide the received discrete time-domain signal into frames of length N; Step 2: For each frame of signal, extract features from multiple dimensions including time domain, frequency domain, time-frequency domain, and statistical domain; Step 3: Feature selection is performed using an optimization strategy that integrates random forest importance evaluation with prior mutation detection correlation weighting. Step 4: Train the classifier; Step 5: Introduce an adaptive decision threshold adjustment mechanism based on maximum a posteriori probability to distinguish between interference and non-interference.

2. The support vector machine interference detection method based on feature selection and maximum a posteriori according to claim 1, characterized in that, In the time domain, mean square power features, peak-to-average power ratio (PAPR) features, and kurtosis are extracted; in the frequency domain, peak power spectral density, maximum spectral peak, spectral entropy, and 3dB bandwidth are extracted; in the time-frequency domain, maximum time-frequency energy and time-frequency spectral entropy are extracted; and in the statistical domain, fractal dimension is calculated.

3. The support vector machine interference detection method based on feature selection and maximum a posteriori according to claim 2, characterized in that, Feature selection is based on random forest importance scoring. Stability score Redundancy penalty factor and cumulative correlation score with CUSUM The obtained weighted importance score The following was carried out: ； in , , Measuring the global discriminative power of features Reflects temporal stability, Adjusting the sensitivity to mutation interference, Control redundancy penalty; sort features by weighted importance score from highest to lowest, and select the highest-scoring feature subset S such that: ； in .

4. The support vector machine interference detection method based on feature selection and maximum a posteriori according to claim 3, characterized in that, The specific process for step 4 is as follows: Let the training sample set after feature selection optimization be: ； in This represents the d-dimensional optimal feature vector of the i-th training sample. These represent the category labels "no interference" and "with interference" respectively, and M is the total number of training sample signal frames; The Gaussian radial basis function kernel is selected as the kernel mapping for the support vector machine: ； in The optimal value for the kernel width parameter is determined through grid search combined with cross-validation. The optimization objective of support vector machines is: ; The constraints are: ; Where w is the weight vector and b is the bias term. For kernel mapping, As a penalty factor, These are slack variables; The sequential minimum optimization algorithm is used to solve this convex quadratic programming problem, yielding the optimal set of support vectors and the corresponding Lagrange multipliers. The final decision function obtained after training: ； The decision function outputs continuous real numbers, with the positive or negative sign indicating the category and the absolute value representing the classification confidence.

5. The support vector machine interference detection method based on feature selection and maximum a posteriori according to claim 4, characterized in that, In step 4, during the training phase, the probability density estimates of the support vector machine output under both the interference and non-interference classes are additionally saved. and .

6. The support vector machine interference detection method based on feature selection and maximum a posteriori according to claim 4, characterized in that, The specific process for step 5 is as follows: The feature vector after feature selection is: The decision function obtained from training the support vector machine is: The probability density functions for the scoring of the interference class and the non-interference class are respectively and The prior probabilities of the samples are respectively and The maximum a posteriori probability decision formula is: ； This judgment is equivalently transformed into a comparison. With the optimal threshold Size; The fitted probability density functions of the interference and non-interference classes are obtained by using kernel density estimation or Gaussian mixture models. and Calculate at the threshold The following are the probabilities of missed detection and false alarms: ； Construction optimization problem: ; in This is a weighting factor used to adjust the importance of missed detections and false alarms in the optimization objective; By solving this optimization problem, we obtain It will adaptively change according to the scoring distribution characteristics of different interference scenarios; The final judgment formula is: ； Complete the distinction between interference and non-interference states.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the support vector machine interference detection method based on feature selection and maximum a posteriori as described in any one of claims 1-6.

8. A computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the support vector machine interference detection method based on feature selection and maximum a posteriori as described in any one of claims 1-6.

Images