Sea surface target detection method based on time-frequency feature enhancement and false alarm rate control

Abstract

The invention relates to the technical field of radar signal processing and target detection, in particular to a sea-surface target detection method based on time-frequency feature enhancement and false alarm rate control, which comprises the following steps of: performing multi-scale decomposition on a radar echo signal by adopting a fully adaptive noise set empirical mode decomposition algorithm, extracting an intrinsic mode component to realize signal reconstruction, and extracting an intrinsic mode component; target features are highlighted; performing short-time Fourier transform on the reconstructed signal and the original signal subjected to moving average filtering processing, and mapping a time-frequency map into an RGB image so as to enhance the characterization capability of the time-frequency feature; a convolutional neural network with a false alarm rate control mechanism is designed, a fused time-frequency image is used as input, and accurate target detection and classification identification are realized; target signals and sea clutter interference are effectively separated, the identifiability of target features is remarkably improved, and the identification threshold can be dynamically adjusted.

Description

Technical Field

[0001] The invention relates to the technical field of radar signal processing and target detection, and in particular to a sea surface target detection method based on time-frequency feature enhancement and false alarm rate control. Background Art

[0002] Radar, as the main means of target detection, tracking and identification, plays an important role in maritime safety, shipping management and military defense. With the continuous advancement of high-resolution radar technology, the accuracy of radar signal detection in complex sea conditions has been improved to a certain extent, promoting its application in marine defense and marine safety. Sea clutter, as an interference signal, contains a large number of spectral components, and its characteristics will change significantly with the changes in external conditions such as wind, waves and sea conditions, which are reflected in non-Gaussian, nonlinear, non-stationary and time-varying characteristics. The existence of sea clutter poses a great challenge to target detection, and weak target signals are often submerged. Traditional target detection algorithms are mainly based on the construction of models based on the amplitude statistical characteristics of sea clutter. However, in practical applications, these models cannot perfectly fit sea clutter, resulting in limited detection performance. In recent years, deep learning has been widely used in the field of radar target detection, which can automatically learn high-level features from data. However, the model is easily affected by complex backgrounds, lacks robustness to clutter interference, and is difficult to adapt to detection needs under different sea conditions. Summary of the invention

[0003] The purpose of the present invention is to provide a sea surface target detection method based on time-frequency feature enhancement and false alarm rate control to solve the problems raised in the above background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0005] A method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control, the method comprising:

[0006] Step 1: Obtain a radar data set consisting of observation data samples and divide it using a sliding window;

[0007] Step 2: Decompose and reconstruct the radar signal of each sample in the radar data set;

[0008] Step 3: Extract and map time-frequency features based on the reconstructed signal;

[0009] Step 4: Divide the radar data set into training set and test set in chronological order;

[0010] Split the data set into a training set and a test set: Since the sliding window is used in step 1 to achieve the division of short-term observation data samples, in order to avoid evaluation errors due to overlapping sample features, the training set and the test set are divided in 8:2 in chronological order;

[0011] Step 5: Design a convolutional neural network-based feature extraction network and false alarm rate control module based on the training set, and build a convolutional neural network model.

[0012] Preferably, step one comprises:

[0013] Obtain a radar data set consisting of observation data samples. The observation time of the selected sample is 1.024s. The radar data set is divided using a sliding window. The sliding window width N = 1024 and the sliding step length Δn = 32. The formula for generating data samples is as follows: i [n]=x(Δn·(i-1)+1:Δn·(i-1)+N).

[0014] Preferably, step 2 comprises:

[0015] S201, obtaining each sample in the radar data set, performing CEEMDAN decomposition on each divided sample, extracting several intrinsic mode functions IMFs, and confirming different frequency components of the signal;

[0016] S202, IMF components obtained based on the intrinsic mode function, according to the formula: The mutual correlation coefficient ρ between the IMF component and the radar echo signal is calculated i , thereby determining the frequency band boundary;

[0017] Among them, cov(c i (t),x(t)) represents the i-th IMF component c i The covariance of (t) and the original signal x(t), and c i The variance of (t) and x(t);

[0018] S203, removing the highest frequency component and trend item: all IMF components before the dividing line are regarded as high frequency components, and the remaining components are regarded as low frequency IMF components;

[0019] S204, suppressing sea clutter interference by reconstructing and enhancing the target signal: performing sliding average filtering on the original signal to smooth the noise signal, and generating a combination of the enhanced signal and the original signal;

[0020] The sliding average filter formula is as follows:

[0021]

[0022] Where M represents the length of the sliding window.

[0023] Preferably, step three includes:

[0024] S301, obtaining the reconstructed low-frequency and high-frequency signals and the filtered signal, according to the formula: Performing short-time Fourier transform on the reconstructed low-frequency and high-frequency signals and the filtered signal to convert the time series signal into a time-frequency domain representation;

[0025] Among them, the sampling frequency f s =500, window length N=64;

[0026] S302 , normalizing the time-frequency spectrogram values ​​of the low-frequency signal, the high-frequency signal, and the sliding average filter signal and mapping them into an RGB three-channel color time-frequency image.

[0027] Preferably, in step 5, constructing a convolutional neural network model includes: a feature extraction network and a false alarm rate control module;

[0028] S501, the feature extraction network is used to extract target features from the time-frequency image and output classification probability;

[0029] S502, the false alarm rate control module includes a threshold calculation logic for dynamically adjusting the detection threshold according to the target false alarm rate to ensure the detection performance of the model under different background complexities;

[0030] According to the false alarm rate dynamic control formula: threshold = O1(i), i = pfa desired ·N clutter ; Customize the classification thresholds of target signals and sea clutter signals to achieve dynamic control of false alarm rate;

[0031] Among them, Ο1 represents the output result of the sea clutter samples after sorting from the first to the smallest arrival, pfa desired Indicates the false alarm rate set, N clutter represents the number of sea clutter samples;

[0032] After being processed by the softmax function, the output value is a two-dimensional vector; the first element represents the probability that the sample is classified as a sea clutter signal, and the second element represents the probability that the sample is classified as a target signal;

[0033] S503, model training and reasoning: using the cross entropy loss function, the model optimizer is set to Adam, and the learning rate is 1e-4; reasoning is performed on the sea clutter samples in the training set, and the classification probability vector is generated using the feature extraction network:

[0034]

[0035] wherein represents the probability that the i-th sample is classified as sea clutter, represents the probability that the i-th sample is classified as a target signal, and N clutter is the number of sea clutter samples in the training set;

[0036] Then, according to the false alarm rate dynamic control formula, the threshold T is obtained; for the test set samples, the classification probability is calculated. If p clutter < T, it is determined as a target signal; if p clutter ≥ T, it is determined as a sea clutter signal.

[0037] Preferably, S501 in step five includes:

[0038] The feature extraction network includes a convolutional feature extraction module, a CBAM module, and a fully connected layer;

[0039] To fully extract the information of different frequency bands corresponding to different channels, depthwise separable convolution is introduced. By convolving each channel separately, the independence of low-frequency, high-frequency, and local dynamic features is retained, avoiding information interference between channels; subsequently, pointwise convolution is used to achieve cross-frequency band feature fusion; the CBAM module can adaptively adjust the importance of each channel and spatial region, strengthening the frequency band and target features.

[0040] A computer-readable storage medium stores a computer program thereon, and when the program is executed by a processor, it implements the steps in the above-mentioned sea surface target detection method based on time-frequency feature enhancement and false alarm rate control.

[0041] A computer device includes a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the program, it implements the steps in the above-mentioned sea surface target detection method based on time-frequency feature enhancement and false alarm rate control.

[0042] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0043] The present application provides a method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control. First, the CEEMDAN algorithm is used to perform multi-scale decomposition of radar signals, extract intrinsic mode components to reconstruct signals, effectively highlight target features, and weaken the influence of sea clutter interference on detection. Secondly, by performing short-time Fourier transform on the reconstructed signal and the filtered signal, the time-frequency spectrum is mapped into three RGB channels, realizing the multi-dimensional representation of target information and significantly improving the feature differentiation capability. In the feature extraction process, based on the convolutional neural network model, the deep separable convolution module is used to efficiently extract features of different frequency bands, and the target area features are further enhanced by combining the CBAM module to improve the network's target detection accuracy and robustness under complex backgrounds. In response to different false alarm rate requirements, the present application designs a false alarm rate control module, which dynamically sorts the classification probabilities of sea clutter samples and calculates the classification threshold, thereby adjusting the classification criteria of target signals and sea clutter signals to meet the preset false alarm rate requirements. The present application realizes small target detection under complex sea clutter backgrounds, overcomes the problems of low detection accuracy and poor adaptability of traditional methods, and is suitable for a variety of complex application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0045] Figure 1 It is a flow chart of a method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control of the present invention;

[0046] Figure 2 It is a structural diagram of the network model of the present invention;

[0047] Figure 3 It is the decomposition result of the real and imaginary parts of a sample by CEEMDAN;

[0048] Figure 4 It is the low-frequency, high-frequency, STFT image after sliding average filtering and three-channel mapping of the target signal;

[0049] Figure 5 It is the low frequency, high frequency, STFT image after sliding average filtering and three-channel mapping of the clutter signal;

[0050] Figure 6 It is a prediction accuracy diagram of the present invention under different false alarm rates on the test set. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1-Figure 6 , the present invention provides a technical solution:

[0053] Embodiment 1:

[0054] A method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control, the method comprising:

[0055] Step 1: Obtain a radar data set consisting of observation data samples and divide it using a sliding window;

[0056] Preferably, step one comprises:

[0057] Obtain a radar data set consisting of observation data samples. The observation time of the selected sample is 1.024s. The radar data set is divided using a sliding window. The sliding window width N = 1024 and the sliding step length Δn = 32. The formula for generating data samples is as follows: i [n]=x(Δn·(i-1)+1:Δn·(i-1)+N).

[0058] Step 2: Decompose and reconstruct the radar signal of each sample in the radar data set;

[0059] Preferably, step 2 comprises:

[0060] S201, obtaining each sample in the radar data set, performing CEEMDAN decomposition on each divided sample, extracting several intrinsic mode functions IMFs, and confirming different frequency components of the signal;

[0061] S202, IMF components obtained based on the intrinsic mode function, according to the formula: The mutual correlation coefficient ρ between the IMF component and the radar echo signal is calculated i , thereby determining the frequency band boundary;

[0062] Among them, cov(c i (t),x(t)) represents the i-th IMF component c i The covariance of (t) and the original signal x(t), and c i The variance of (t) and x(t);

[0063] S203, removing the highest frequency component and trend item: all IMF components before the dividing line are regarded as high frequency components, and the remaining components are regarded as low frequency IMF components;

[0064] S204, suppressing sea clutter interference by reconstructing and enhancing the target signal: performing sliding average filtering on the original signal to smooth the noise signal, and generating a combination of the enhanced signal and the original signal;

[0065] The sliding average filter formula is as follows:

[0066]

[0067] Where M represents the length of the sliding window.

[0068] Step 3: Extract and map time-frequency features based on the reconstructed signal;

[0069] Preferably, step three includes:

[0070] S301, obtaining the reconstructed low-frequency and high-frequency signals and the filtered signal, according to the formula: Performing short-time Fourier transform on the reconstructed low-frequency and high-frequency signals and the filtered signal to convert the time series signal into a time-frequency domain representation;

[0071] Among them, the sampling frequency f s =500, window length N=64;

[0072] S302 , normalizing the time-frequency spectrogram values ​​of the low-frequency signal, the high-frequency signal, and the sliding average filter signal and mapping them into an RGB three-channel color time-frequency image.

[0073] Step 4: Divide the radar data set into training set and test set in chronological order;

[0074] Split the data set into a training set and a test set: Since the sliding window is used in step 1 to achieve the division of short-term observation data samples, in order to avoid evaluation errors due to overlapping sample features, the training set and the test set are divided in 8:2 in chronological order;

[0075] Step 5: Design a feature extraction network and a false alarm rate control module based on a convolutional neural network based on the training set, and build a convolutional neural network model;

[0076] Preferably, in step 5, constructing a convolutional neural network model includes: a feature extraction network and a false alarm rate control module;

[0077] S501, the feature extraction network is used to extract target features from the time-frequency image and output classification probability;

[0078] Preferably, S501 in step five includes:

[0079] The feature extraction network includes a convolutional feature extraction module, a CBAM module, and a fully connected layer;

[0080] To fully extract the information of different frequency bands corresponding to different channels, depthwise separable convolution is introduced. By convolving each channel separately, the independence of low-frequency, high-frequency, and local dynamic features is retained, avoiding information interference between channels; subsequently, pointwise convolution is used to achieve cross-frequency band feature fusion; the CBAM module can adaptively adjust the importance of each channel and spatial region, strengthening the frequency band and target features.

[0081] S502. The false alarm rate control module contains threshold calculation logic for dynamically adjusting the detection threshold according to the target false alarm rate to ensure the detection performance of the model under different background complexities;

[0082] According to the dynamic control formula of the false alarm rate: threshold = O1(i), i = pfa desired ·N clutter ; Customize the classification threshold of the target signal and the sea clutter signal to achieve dynamic control of the false alarm rate;

[0083] Among them, Ο1 represents the output result after sorting the sea clutter samples in ascending order according to the first digit, and pfa desired represents the set false alarm rate, and N clutter represents the number of sea clutter samples;

[0084] After being processed by the softmax function, the output value is a two-dimensional vector; the first element represents the probability that the sample is classified as a sea clutter signal, and the second element represents the probability that the sample is classified as a target signal;

[0085] S503. Model training and inference: The cross-entropy loss function is adopted, the model optimizer is set to Adam, and the learning rate is 1e-4; the sea clutter samples in the training set are inferred, and the feature extraction network is used to generate a classification probability vector:

[0086]

[0087] Among them represents the probability that the i-th sample is classified as sea clutter, represents the probability that the i-th sample is classified as a target signal, and N clutter is the number of sea clutter samples in the training set;

[0088] Then, according to the dynamic control formula of the false alarm rate, the threshold T is obtained; for the test set samples, calculate the classification probability. If p clutter <T, it is determined as a target signal; if p clutterIf it is ≥T, it is determined as sea clutter signal.

[0089] Based on this, the short-time observation data samples of the radar dataset are divided. The observation time is set to 1.024 seconds, the sliding step size is set to 32 sampling points, and a total of 44715 samples are generated. The real and imaginary parts of each sample are respectively decomposed by CEEMDAN to obtain multiple intrinsic mode components. Calculate the correlation coefficient between each IMF component and the original signal, determine the frequency band boundary line, and the boundary line of the present invention is 6. Remove the highest frequency component and the trend term, reconstruct the signal for the remaining components, and classify the low-frequency components and high-frequency components separately. Use moving average filtering to smooth the original signal, and combine the reconstructed signal to generate an enhanced signal. Perform short-time Fourier transform on the enhanced signal respectively, set the sampling frequency to 500 and the window length to 64 to obtain the time-frequency domain representation, and the size of the time-frequency diagram is 65×64. Map the STFT results of the low-frequency signal, high-frequency signal, and smoothed signal to the RGB channels to synthesize a color time-frequency image to enhance the characterization ability of target features. Divide the dataset according to the ratio of 8:2, with the first 80% of the samples as the training set and the last 20% of the samples as the test set. There are 3252 target samples and 32520 clutter samples in the training set, and 813 target samples and 8130 clutter samples in the test set.

[0090] Build a convolutional neural network model with controllable false alarm rate according to Step 5 and train the network. The optimizer of the network model is set to Adam, the initial learning rate is 1e-4, and the number of training epochs is 50. Dynamically adjust the detection threshold T according to the target false alarm rate requirement, calculate the classification probability P for the test set samples, if P<T, it is determined as the target signal, otherwise it is determined as the sea clutter signal.

[0091] After 50 rounds of training, when the false alarm rate is 0.1, the accuracy of the test set is 0.9987; when the false alarm rate is 0.01, the accuracy is 0.9876; when the false alarm rate is 0.001, the accuracy is 0.9089.

[0092] Embodiment 2:

[0093] The computer-readable storage medium of this embodiment stores a computer program, and when the program is executed by a processor, it implements the steps in a sea surface target detection method based on time-frequency feature enhancement and false alarm rate control in Embodiment 1.

[0094] The computer-readable storage medium of this embodiment can be the internal storage unit of the terminal, such as the hard disk or memory of the terminal; the computer-readable storage medium of this embodiment can also be the external storage device of the terminal, such as the plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the terminal.

[0095] 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 is to be output.

[0096] Embodiment 3:

[0097] The computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of a method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control in Embodiment 1 are implemented.

[0098] In this embodiment, the processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, readily available programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0099] Those skilled in the art will appreciate that the disclosed content of the embodiments may be provided as methods, systems, or computer program products. Therefore, the present solution may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present solution may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

[0100] The present solution is described with reference to the method according to the embodiment of the present solution and the flowchart and / or block diagram of the computer program product. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions; these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 one or more processes and / or methods Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0101] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 one or more processes and / or methods Figure 1 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 one or more processes and / or methods Figure 1 The steps for the functions specified in one or more boxes.

[0103] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0104] Finally, it should be noted that the above description is only 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 aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A sea surface target detection method based on time-frequency feature enhancement and false alarm rate control, characterized in that: The method comprises: Step 1: Obtain a radar data set consisting of observation data samples and divide it using a sliding window; Step 2: Decompose and reconstruct the radar signal of each sample in the radar data set; Step 3: Extract and map time-frequency features based on the reconstructed signal; Step 4: Divide the radar data set into training set and test set in chronological order; Step 5: Design a convolutional neural network-based feature extraction network and false alarm rate control module based on the training set, and build a convolutional neural network model.

2. A method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as claimed in claim 1, characterized in that: The step one comprises: Obtain a radar data set consisting of observation data samples. The observation time of the selected sample is 1.024s. The radar data set is divided using a sliding window. The sliding window width N = 1024 and the sliding step length Δn = 32. The formula for generating data samples is as follows: i [n]=x(Δn·(i-1)+1:Δn·(i-1)+N).

3. The method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as claimed in claim 1, characterized in that: The step 2 comprises: S201, obtaining each sample in the radar data set, performing CEEMDAN decomposition on each divided sample, extracting several intrinsic mode functions IMFs, and confirming different frequency components of the signal; S202, IMF components obtained based on the intrinsic mode function, according to the formula: The mutual correlation coefficient ρ between the IMF component and the radar echo signal is calculated i , thereby determining the frequency band boundary; Among them, cov(c i (t),x(t)) represents the i-th IMF component c i The covariance of (t) and the original signal x(t), and c i The variance of (t) and x(t); S203, removing the highest frequency component and trend item: all IMF components before the dividing line are regarded as high frequency components, and the remaining components are regarded as low frequency IMF components; S204, suppressing sea clutter interference by reconstructing and enhancing the target signal: performing sliding average filtering on the original signal to smooth the noise signal, and generating a combination of the enhanced signal and the original signal; The sliding average filter formula is as follows: Where M represents the length of the sliding window.

4. The method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as claimed in claim 1, characterized in that: The step three comprises: S301, obtaining the reconstructed low-frequency and high-frequency signals and the filtered signal, according to the formula: Performing short-time Fourier transform on the reconstructed low-frequency and high-frequency signals and the filtered signal to convert the time series signal into a time-frequency domain representation; Among them, the sampling frequency f s =500, window length N=64; S302 , normalizing the time-frequency spectrogram values ​​of the low-frequency signal, the high-frequency signal, and the sliding average filter signal and mapping them into an RGB three-channel color time-frequency image.

5. The method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as claimed in claim 1, characterized in that: In the step 5, constructing a convolutional neural network model includes: a feature extraction network and a false alarm rate control module; S501, the feature extraction network is used to extract target features from the time-frequency image and output classification probability; S502, the false alarm rate control module includes a threshold calculation logic for dynamically adjusting the detection threshold according to the target false alarm rate to ensure the detection performance of the model under different background complexities; According to the false alarm rate dynamic control formula: threshold = O1(i), i = pfa desired ·N clutter ; Customize the classification thresholds of target signals and sea clutter signals to achieve dynamic control of false alarm rate; Among them, Ο1 represents the output result of the sea clutter samples after sorting from the first to the smallest arrival, pfa desired Indicates the false alarm rate set, N clutter represents the number of sea clutter samples; After being processed by the softmax function, the output value is a two-dimensional vector; the first element represents the probability that the sample is classified as a sea clutter signal, and the second element represents the probability that the sample is classified as a target signal; S503, model training and reasoning: using the cross entropy loss function, the model optimizer is set to Adam, and the learning rate is 1e-4; reasoning is performed on the sea clutter samples in the training set, and the classification probability vector is generated using the feature extraction network: in represents the probability that the i-th sample is classified as sea clutter, represents the probability that the i-th sample is classified as the target signal, N clutter is the number of sea clutter samples in the training set; Then, according to the false alarm rate dynamic control formula, the threshold T is obtained; for the test set samples, the classification probability is calculated. If p clutter < T, it is determined as the target signal; if p clutter ≥ T, it is determined as the sea clutter signal.

6. A method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as claimed in claim 5, characterized in that: S501 in the step 5 includes: The feature extraction network includes a convolutional feature extraction module, a CBAM module and a fully connected layer; In order to fully extract the information of different frequency bands corresponding to different channels, deep separable convolution is introduced. By performing convolution on each channel separately, the independence of low-frequency, high-frequency and local dynamic features is retained to avoid information interference between channels. Subsequently, cross-band feature fusion is achieved through point-by-point convolution. The CBAM module can adaptively adjust the importance of each channel and spatial area to enhance frequency band and target features.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a method for detecting sea surface targets based on time-frequency feature enhancement and false alarm rate control as described in any one of claims 1 to 6 are implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps in the sea surface target detection method based on time-frequency feature enhancement and false alarm rate control as described in any one of claims 1-6 are implemented.

Images