Underwater target identification method and device, computer equipment and readable storage medium

Abstract

The embodiment of the invention provides an underwater target identification method and device, computer equipment and a readable storage medium. The embodiment of the invention provides an underwater target identification method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring an initial echo signal reflected by an underwater target aiming at a sonar signal, and generating an echo time-frequency graph; determining a time-frequency sub-image corresponding to at least one micro-motion period according to the echo time-frequency graph; projecting the time-frequency sub-images in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-images, and constructing an energy distribution image based on the multiple projection intensity distributions; obtaining a sine curve sequence corresponding to the micro-Doppler component in the time-frequency curve according to the energy distribution image; micro-Doppler components are extracted based on the sine curve sequence, and a micro-Doppler time-frequency diagram is generated; and inputting the micro-Doppler time-frequency graph into a pre-trained classification model to obtain an underwater target identification result. Therefore, the accuracy of underwater target identification can be improved.

Description

Technical Field

[0001] The present application relates to the field of underwater acoustic detection technology, and in particular to an underwater target recognition method, device, computer equipment and readable storage medium. Background Art

[0002] The ocean covers more than 70% of the earth's surface, about 360 million square kilometers, and is an important strategic space for sustainable development of mankind. Developing marine resources and developing marine science and technology have become a global trend. As an important part of marine information perception, underwater target detection technology has played an important role in marine science, marine security and marine economy, and has great strategic value.

[0003] In the related art, since sound waves have good propagation performance underwater, underwater acoustic communication can be carried out through sound waves as carriers, and target detection can be achieved based on the extracted underwater acoustic signals. However, since the underwater acoustic channel is a complex channel with characteristics such as severe noise and multipath interference, and the acoustic energy loss caused by the absorption loss of seawater, the reflection loss of the boundary, and the expansion loss of sound waves, the signal quality is not high and the signal interference is high, which reduces the accuracy of underwater target recognition. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose an underwater target recognition method, device, computer equipment and readable storage medium, which can improve the accuracy of underwater target recognition.

[0005] To achieve the above object, a first aspect of an embodiment of the present application proposes an underwater target recognition method, the method comprising: Acquire an initial echo signal reflected by an underwater target in response to a sonar signal, and generate an echo time-frequency diagram based on the initial echo signal; Determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; Projecting the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructing an energy distribution image based on the multiple projection intensity distributions; Wherein, there are overlapping projections between different projection intensity distributions in the energy distribution image; Determine a target energy region having the highest cumulative projection intensity value from the energy distribution image, and obtain a sinusoidal curve sequence corresponding to a micro-Doppler component in the time-frequency curve based on a target position of the target energy region in the energy distribution image; Extracting a micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generating a micro-Doppler time-frequency diagram according to the micro-Doppler component; The micro-Doppler time-frequency diagram is input into a pre-trained classification model to obtain a recognition result of the underwater target.

[0006] Accordingly, a second aspect of an embodiment of the present application provides an underwater target recognition device, the device comprising: An acquisition module, used for acquiring an initial echo signal reflected by an underwater target in response to a sonar signal, and generating an echo time-frequency diagram based on the initial echo signal; A first determination module is used to determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; A construction module, used to project the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and construct an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image; A second determination module is used to determine a target energy region with the highest cumulative projection intensity value from the energy distribution image, and obtain a sinusoidal curve sequence corresponding to a micro-Doppler component in the time-frequency curve based on a target position of the target energy region in the energy distribution image; A generating module, configured to extract a micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generate a micro-Doppler time-frequency diagram according to the micro-Doppler component; An input module is used to input the micro-Doppler time-frequency diagram into a pre-trained classification model to obtain a recognition result of the underwater target.

[0007] In some implementations, the acquisition module is further used to: Acquiring a radial displacement velocity of the underwater target, a sound velocity of an underwater environment corresponding to the underwater target, and a carrier frequency of the sonar signal, and determining a predicted Doppler frequency of the underwater target based on the radial displacement velocity, the sound velocity, and the carrier frequency; Obtaining a preset filter range coefficient, and setting upper and lower cutoff frequencies of a bandpass filter based on the carrier frequency, the filter range coefficient and the predicted Doppler frequency; The initial echo signal is filtered using the upper and lower cutoff frequencies to obtain a first echo signal; performing normalization processing on the first echo signal to obtain a second echo signal; Perform time-frequency transformation on the second echo signal to obtain an echo time-frequency diagram corresponding to the initial echo signal.

[0008] In some embodiments, the underwater target recognition device further includes a processing module for: Sequentially intercepting signals to be processed from the first echo signal through a preset sliding window; For the signal to be processed corresponding to each sliding window, obtain the mean and standard deviation of the signal to be processed, and set a threshold range of the signal to be processed based on the mean and the standard deviation; When a target signal value in the signal to be processed is outside the threshold range, replacing the target signal value with the mean value; After the processing of the multiple signals to be processed contained in the first echo signal is completed, a processed first echo signal is obtained.

[0009] In some implementations, the first determining module is further configured to: Performing autocorrelation analysis on the echo time-frequency diagram to obtain an autocorrelation function corresponding to the initial echo signal; wherein the autocorrelation function is used to measure the degree of overlap of the time-frequency curves under different time delays; Obtaining a first target peak value of the autocorrelation function; Based on the position of the first target peak, the micro-motion period of the underwater target is determined.

[0010] In some embodiments, the underwater target recognition device further includes an adjustment module for: Performing a linear transformation on the time-frequency sub-image to obtain a linear component in the time-frequency sub-image; For each straight line component, obtain the slope and intercept of each straight line component, and determine the second-order translation result of the straight line component based on the slope and the intercept; When the second-order translation result indicates that the linear component has a second-order translation, the frequency axis coordinates in the time-frequency sub-image are offset-adjusted based on the slope and the intercept to obtain a processed time-frequency sub-image.

[0011] In some embodiments, the underwater target recognition device further includes a training module for: Acquire multiple first echo time-frequency diagrams corresponding to the underwater target in the actual environment; the first echo time-frequency diagrams correspond to multiple micro-motion forms; By using a preset diffusion model, sample expansion is performed based on a plurality of first echo time-frequency diagrams to obtain a plurality of corresponding second echo time-frequency diagrams; Extracting micro-Doppler components from the first echo time-frequency diagram and the second echo time-frequency diagram respectively to obtain a plurality of sample micro-Doppler time-frequency diagrams, and generating a sample training set based on the plurality of sample micro-Doppler time-frequency diagrams; Inputting each sample micro-Doppler time-frequency graph in the sample training set into a preset classification model in turn to obtain a predicted category label corresponding to each sample data; Obtaining a sample category label of the sample data, and determining a target loss based on a difference between the predicted category label and the sample category label; Based on the target loss, the parameters of the preset classification model are adjusted to obtain a classification model.

[0012] In some embodiments, the training module is further used to: For each first echo time-frequency diagram, converting the first echo time-frequency diagram into a one-dimensional time series; The one-dimensional time series is input into a preset diffusion model so that the diffusion model is expanded based on the first echo time-frequency diagram to obtain a plurality of second echo time-frequency diagrams having the same micro-motion form as the first echo time-frequency diagram; wherein the first dimension of each convolution kernel in the diffusion model is adapted to the dimension of the one-dimensional time series.

[0013] Correspondingly, the third aspect of the embodiments of the present application proposes a computer device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the underwater target recognition method described in any one of the embodiments of the first aspect of the present application when executing the computer program.

[0014] Correspondingly, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the underwater target recognition method described in any one of the embodiments of the first aspect of the present application.

[0015] The embodiment of the present application obtains the initial echo signal reflected by the underwater target in response to the sonar signal, and generates an echo time-frequency diagram based on the initial echo signal; determines the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determines at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; projects the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructs an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image; determines the target energy area with the highest cumulative projection intensity value from the energy distribution image, and obtains a sinusoidal curve sequence corresponding to the micro-Doppler component in the time-frequency curve based on the target position of the target energy area in the energy distribution image; extracts the micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generates a micro-Doppler time-frequency diagram based on the micro-Doppler component; inputs the micro-Doppler time-frequency diagram into a pre-trained classification model to obtain the recognition result of the underwater target. In this way, the micro-motion period of the underwater target can be determined from the echo time-frequency diagram to better understand the inherent motion characteristics of the underwater target, thereby improving the accuracy of recognition. In addition, the main distribution direction of the micro-Doppler feature is consistent, and it will appear more concentrated in the projected space. By determining the target position of the target energy area with the highest cumulative projection intensity in the energy distribution image, the sinusoidal curve sequence corresponding to the micro-Doppler component can be accurately determined to accurately separate the micro-Doppler component corresponding to the underwater target from the complex echo signal for identification and classification by the classification model, providing the classification model with clearer and more representative features, effectively reducing the impact of interference signals, and improving the accuracy of recognition. In summary, the present application can improve the accuracy of underwater target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the architecture of the underwater target recognition system provided by an embodiment of the present application; Figure 2 is a flow chart of an underwater target recognition method provided by an embodiment of the present application; Figure 3 is a time-frequency sub-image reconstructed after the inverse Radon transform provided in the embodiment of the present application; Figure 4 is a diagram of steps for extracting micro-Doppler components provided in an embodiment of the present application; Figure 5 It is a schematic diagram of the micro-motion form provided in the embodiment of the present application; Figure 6 This is a flow chart of a classification model obtained through training provided in an embodiment of the present application; Figure 7 is a structural diagram of a diffusion model provided in an embodiment of the present application; Figure 8It is a schematic diagram of the functional modules of the underwater target recognition device provided in an embodiment of the present application; Fig. 9 It is a schematic diagram of the hardware structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0018] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0020] The ocean covers more than 70% of the earth's surface, about 360 million square kilometers, and is an important strategic space for sustainable development of mankind. Developing marine resources and developing marine science and technology have become a global trend. As an important part of marine information perception, underwater target detection technology has played an important role in marine science, marine security and marine economy, and has great strategic value.

[0021] In the related art, since sound waves have good propagation performance underwater, underwater acoustic communication can be carried out through sound waves as carriers, and target detection can be achieved based on the extracted underwater acoustic signals. However, since the underwater acoustic channel is a complex channel with characteristics such as severe noise and multipath interference, and the acoustic energy loss caused by the absorption loss of seawater, the reflection loss of the boundary, and the expansion loss of sound waves, the signal quality is not high and the signal interference is high, which reduces the accuracy of underwater target recognition.

[0022] Based on this, the embodiments of the present application provide an underwater target recognition method, apparatus, computer equipment and readable storage medium, which can improve the accuracy of underwater target recognition.

[0023] The underwater target recognition method, device, computer equipment and readable storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the underwater target recognition system in the embodiments of the present application is described.

[0024] Please refer to Figure 1 In some implementations, an embodiment of the present application provides an underwater target recognition system, including a terminal 11 and a server 12.

[0025] Exemplarily, terminal 11 can be a sonar sensor, an autonomous underwater vehicle, a data acquisition system, etc. Terminal 11 can interact with the underwater environment, be responsible for sending sonar signals, and collect initial echo signals reflected by underwater targets in response to sonar signals, as well as perform preliminary processing on the initial echo signals, such as filtering, processing of outliers, etc.

[0026] Furthermore, the server end 12 may be a computer device such as a high-performance computing server, a cloud server or a data center. The server end 12 may receive the initial echo signal sent by the terminal 11, and process the initial echo signal such as time-frequency analysis, feature extraction, etc., and run a classification model to identify underwater targets.

[0027] The underwater target recognition method in the embodiment of the present application can be illustrated by the following embodiment.

[0028] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first. Moreover, the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or separate consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or separate consent, the necessary user-related data for enabling the normal operation of the embodiment of the present application will be obtained.

[0029] In the embodiment of the present application, the underwater target recognition device will be described from the perspective of the underwater target recognition device, which can be integrated into a computer device. Figure 2 , Figure 2 This is a flowchart of the steps of the underwater target recognition method provided in the embodiment of the present application. In the embodiment of the present application, the underwater target recognition device is specifically integrated in a terminal or a server as an example. When the processor on the terminal or the server executes the program instructions corresponding to the underwater target recognition method, the specific process is as follows: Step 101: Acquire an initial echo signal reflected by an underwater target in response to a sonar signal, and generate an echo time-frequency diagram based on the initial echo signal.

[0030] In some embodiments, in order to facilitate the classification and identification of underwater targets, a sonar signal can be sent to the underwater target. After receiving the initial echo signal reflected by the underwater target, the complex initial echo signal (time domain signal) is converted into an intuitive time-frequency domain representation to provide a basis for accurate classification and identification of the underwater target.

[0031] Underwater targets can be various objects or entities in the water environment, such as natural targets such as marine life, as well as artificial targets such as ships and submarines, torpedoes, underwater robots, underwater facilities (such as engines, rotor blades, etc.). Different underwater targets have their own unique micro-motion forms (such as vibration, rotation, tumbling, and cone motion), which will form unique micro-Doppler features in the echo signal to facilitate the identification of underwater targets.

[0032] The sonar signal may be a sound wave signal emitted by a sonar device for detecting underwater targets. It can be expressed in the form of a sine wave, for example ,in is the amplitude, is the carrier frequency of the sonar signal, and t is the time.

[0033] The initial echo signal may be the original sonar signal reflected by the underwater target, which contains various information about the underwater target, including but not limited to the physical characteristics and dynamic behavior information of the underwater target, such as the size, shape, material, motion state, etc. of the underwater target. It can be expressed as ,in, is the amplitude of the reflected signal, is the carrier frequency of the sonar signal, is the Doppler frequency of the initial echo signal, ,in, is the Doppler frequency generated by the translation of the underwater target, The micro-Doppler frequency generated for a part of an underwater target.

[0034] Among them, the echo time-frequency diagram can be a two-dimensional image generated by time-frequency transformation (such as short-time Fourier transform) after time-frequency analysis of the initial echo signal, which shows the frequency components of the signal at different time points and can facilitate more intuitive observation and analysis of micro-Doppler characteristics.

[0035] In some embodiments, a sonar device may be used to transmit a sonar signal to an underwater target and collect an initial echo signal reflected from the underwater target. After receiving the initial echo signal, the received echo signal may be preliminarily processed, including bandpass filtering, normalization, etc., to remove noise and interference. Further, a short-time Fourier transform may be used to perform time-frequency analysis on the preprocessed echo signal to obtain an echo time-frequency diagram.

[0036] By acquiring the initial echo signal and generating the corresponding echo time-frequency diagram, the characteristics of the underwater target can be effectively extracted and analyzed, providing an important basis for the subsequent identification and classification of the underwater target.

[0037] In some embodiments, in order to improve the accuracy and reliability of the analysis, the initial echo signal may be bandpass filtered and normalized to effectively suppress noise, and the amplitudes of all signals may be scaled to the same scale to facilitate comparison and analysis and optimize the use of computing resources. For example, the "generating an echo time-frequency diagram based on the initial echo signal" in step 101 may include: (101.1) obtaining a radial displacement velocity of an underwater target, a sound velocity of an underwater environment corresponding to the underwater target, and a carrier frequency of a sonar signal, and determining a predicted Doppler frequency of the underwater target based on the radial displacement velocity, the sound velocity, and the carrier frequency; (101.2) Obtaining a preset filter range coefficient, and setting upper and lower cutoff frequencies of the bandpass filter based on the carrier frequency, the filter range coefficient and the predicted Doppler frequency; (101.3) filtering the initial echo signal through the upper and lower cutoff frequencies to obtain the first echo signal; (101.4) performing normalization processing on the first echo signal to obtain a second echo signal; (101.5) Perform time-frequency transformation on the second echo signal to obtain an echo time-frequency diagram corresponding to the initial echo signal.

[0038] The radial displacement velocity may be a velocity component of an underwater target relative to a sonar device along a direction of sound wave propagation, which determines the magnitude of a Doppler frequency shift.

[0039] The speed of sound may be the propagation speed of sound waves in an underwater environment.

[0040] The carrier frequency may be the basic frequency of the sound wave signal emitted by the sonar device.

[0041] The predicted Doppler frequency may be a range of echo signal frequency changes caused by the movement of the target.

[0042] The filter range coefficient may be a proportional factor for determining the upper and lower cutoff frequencies of the bandpass filter, which determines the bandwidth range of the filter and can be set according to actual conditions.

[0043] The upper and lower cut-off frequencies may be the lowest frequency and the highest frequency allowed to pass through by the bandpass filter.

[0044] The first echo signal may be an initial echo signal that has been subjected to bandpass filtering, from which low-frequency noise and other interference signals are removed.

[0045] The second echo signal may be a result of normalizing the first echo signal, and the normalization operation may make the signal amplitude uniform within a specific range (eg, -0.5 to 0.5).

[0046] For example, the predicted Doppler frequency can be calculated by the ratio of twice the product of the radial displacement velocity and the carrier frequency to the speed of sound. For example, the specific calculation formula can be: ,in, represents the radial displacement velocity, represents the carrier frequency of the sonar signal, Indicates the speed of sound.

[0047] In some embodiments, the initial echo signal can be input into a bandpass filter with designed upper and lower cutoff frequencies for processing, so as to retain the components within a specific frequency range in the initial echo signal and remove noise or interference of other frequencies, thereby obtaining a clearer first echo signal that focuses more on the target features, which is convenient for subsequent analysis and processing. For example, the upper and lower cutoff frequencies of the bandpass filter can be set to .in, represents the carrier frequency of the sonar signal, is the filter range coefficient, Represents the estimated Doppler frequency.

[0048] In some embodiments, the cutoff frequency of the bandpass filter can be adjusted according to the actual signal-to-noise ratio of the sonar signal. For example, when high-frequency noise (greater than a preset frequency threshold) is detected, the filter range coefficient α is automatically reduced to suppress interference; in a low-noise environment (less than a preset noise threshold), α is expanded to retain more valid signals. The scale of the adjusted filter range coefficient α can be set according to actual conditions, and the embodiments of the present application are not limited to this.

[0049] Further, in order to unify the characteristic scale, the first echo signal may be normalized to ensure the accuracy of the calculation. Exemplarily, the normalization formula may be: .in, It can be the normalized result, can be a single data point in the initial echo signal to be normalized, is the maximum value in the initial echo signal, is the minimum value in the initial echo signal. By normalizing each data point in the first echo signal, a second echo signal after the first echo signal is processed can be obtained.

[0050] For example, the second echo signal can be converted from the time domain to the time-frequency domain by time-frequency transformation (such as short-time Fourier transform) to generate an intuitive two-dimensional image, that is, an echo time-frequency diagram. The horizontal axis in the echo time-frequency diagram can represent time (seconds) and the vertical axis represents frequency (Hz).

[0051] Through the above methods, the noise and interference in the signal can be effectively filtered out, and the key frequency part of the signal can be retained; in addition, the normalization process ensures the consistency of the signal feature scale and improves the accuracy of subsequent analysis. Finally, the time-frequency transform converts the processed second echo signal into an echo time-frequency diagram, providing an intuitive two-dimensional image for further analysis and identification of underwater targets.

[0052] In some implementations, in order to improve the overall stability of the signal and avoid distortion of subsequent processing results due to individual outliers, outliers in the first echo signal may be identified and processed to effectively smooth the signal, reduce the impact of mutation values, and improve the quality of the signal. For example, before (101.4), that is, before "normalizing the first echo signal to obtain the second echo signal", it may also include: (A.1) intercepting the signals to be processed in sequence from the first echo signal through a preset sliding window; (A.2) for each signal to be processed corresponding to the sliding window, obtaining the mean and standard deviation of the signal to be processed, and setting a threshold range of the signal to be processed based on the mean and standard deviation; (A.3) When the target signal value in the signal to be processed is outside the threshold range, the target signal value is replaced by the mean value; (A.4) After the processing of the multiple signals to be processed contained in the first echo signal is completed, a processed first echo signal is obtained.

[0053] The sliding window may be a time window of fixed length, and the sliding window may slide sequentially from the beginning to the end of the signal, intercepting a fixed length signal segment each time. The size and step length of the sliding window may be adjusted according to specific needs.

[0054] The signal to be processed may be each first echo signal segment intercepted in the sliding window, and these segments are the basis for subsequent mean value, standard deviation calculation and threshold range setting.

[0055] The standard deviation may be a statistic used to measure the degree of dispersion of the signal to be processed, indicating the degree of deviation of the data relative to its mean value.

[0056] The threshold range may be a range interval for identifying abnormal values, and signal values ​​exceeding the threshold range may be determined to be abnormal.

[0057] The target signal value may be a signal value in the signal to be processed that is outside the corresponding threshold range.

[0058] For example, the threshold range may be , where μ is the mean of multiple signal values ​​in the signal to be processed, σ is the standard deviation of multiple signal values ​​in the signal to be processed, k is a preset coefficient (such as k=3 means three times the standard deviation, and the value of k can be set according to actual conditions), and the target signal value beyond the threshold range is regarded as an outlier or mutation point.

[0059] In a certain implementation, the first echo signal can be divided into multiple smaller segments in turn by a sliding window to obtain multiple signals to be processed intercepted by the sliding window. Afterwards, for each signal to be processed, the mean and standard deviation of all signal values ​​contained in the signal to be processed can be calculated, and the threshold range of the corresponding signal to be processed can be set according to the mean and standard deviation. For example, after calculating the mean μ and standard deviation σ of the signal to be processed 1, if k is 3, the threshold range of the signal to be processed 1 can be determined to be .

[0060] In some embodiments, signal values ​​that exceed a threshold range may be replaced with a mean value, thereby effectively smoothing the signal and reducing the impact of sudden changes in values.

[0061] For example, in addition to determining the threshold range by the mean and standard deviation, the threshold range may also be determined by the percentile method, median absolute deviation (calculating the absolute deviation of each data point from the median and using a certain percentile of the deviation as the value), and the like.

[0062] Exemplarily, the first echo signal may not be divided, and the overall mean and overall standard deviation of all signal values ​​in the first echo signal may be directly obtained, and the threshold range may be determined based on the overall mean and the overall standard deviation. When there is a signal value in the first echo signal that is outside the threshold range, the signal value is replaced with the overall mean. In this way, the efficiency of processing abnormal values ​​of the first echo signal can be improved.

[0063] By dividing the first echo signal into multiple signals to be processed, the signal can be processed in smaller local signal segments to more accurately identify and process local features, to adapt to signal characteristics and application requirements at different stages, and to improve the accuracy and robustness of signal processing.

[0064] Step 102: determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram.

[0065] In some embodiments, in order to focus on key features of the target and reduce interference from irrelevant information, the micro-motion period of the underwater target can be determined, and the time-frequency sub-image corresponding to the micro-motion period can be determined to provide a basis for subsequent micro-Doppler feature extraction.

[0066] The time-frequency curve can be a track in the echo time-frequency diagram that shows the change of frequency components over time, and is used to show the distribution of frequency components of the signal at different time points. The time-frequency curve can be generated from the initial echo signal through time-frequency transformation (such as short-time Fourier transform).

[0067] The micromotion period can be the time required for a complete cycle of a specific micromotion form of an underwater target (such as vibration, rotation, etc.). For example, for simple harmonic vibration, the micromotion period can be determined by autocorrelation analysis or spectrum analysis, and the micromotion period is a key parameter for identifying and extracting micro-Doppler features.

[0068] The time-frequency sub-image can be a small part of the image intercepted from the complete echo time-frequency image, usually corresponding to one or more complete micro-motion cycles. The time-frequency sub-image contains the main information of the target micro-motion characteristics, which is convenient for further processing and analysis.

[0069] In some embodiments, an autocorrelation analysis may be performed on the time-frequency curve in the echo time-frequency diagram to estimate its micromotion period; or, the periodic characteristics of the time-frequency curve may be determined by spectrum analysis, for example, by finding the significant peak position in the spectrum diagram to determine the micromotion period. Then, the micromotion period of the underwater target is estimated based on the results of the autocorrelation analysis or spectrum analysis.

[0070] Furthermore, since the Doppler effect caused by micromotion appears as a periodic feature in the time-frequency diagram, a complete micromotion cycle contains all the key information of the effect, such as the pattern and amplitude of frequency change. Therefore, in order to reduce the complexity and computational complexity of data processing and avoid redundancy and confusion during feature extraction, one or more target time periods corresponding to complete micromotion cycles can be captured in the echo time-frequency diagram, and a time-frequency sub-image corresponding to the target time period can be captured from the echo time-frequency diagram. Each time-frequency sub-image should contain a complete micromotion cycle for subsequent processing.

[0071] By determining the micro-motion cycle of the underwater target and determining the time-frequency sub-image based on the micro-motion cycle, a single micro-motion cycle can be analyzed intensively to improve the accuracy of feature extraction and to more accurately capture the micro-motion features without interference from other cycles.

[0072] In some embodiments, in order to provide an accurate time reference for subsequent micro-Doppler feature extraction and classification and identification, the initial echo signal may be subjected to autocorrelation analysis to accurately identify the periodic features in the initial echo signal, thereby facilitating identification of the micro-motion period of the underwater target. For example, the step 102 of "determining the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram" may include: (102.1) performing autocorrelation analysis on the echo time-frequency diagram to obtain an autocorrelation function corresponding to the initial echo signal; wherein the autocorrelation function is used to measure the degree of overlap of the time-frequency curve under different time delays; (102.2) Obtain the first target peak of the autocorrelation function; (102.3) Determine the micromotion period of the underwater target based on the location of the first target peak.

[0073] Among them, the autocorrelation function can be used to measure the similarity or overlap of signals at different time delays.

[0074] Among them, the first target peak can be the first peak position in the autocorrelation function that is significantly higher than the background noise level, which corresponds to the main periodic component of the signal, that is, the micro-motion period of the target, and is the key reference point for determining the micro-motion period.

[0075] Furthermore, assuming that the time-frequency curve is , its autocorrelation function It can be expressed as: ; in, is the time delay, by calculating different time delays Next , the autocorrelation function can be obtained.

[0076] Furthermore, since the first target peak of the autocorrelation function often corresponds to the periodic characteristics of the time-frequency curve, the autocorrelation function can be determined by analyzing the graph of the autocorrelation function. The position of the first target peak in (Apart from The first target peak indicates that the time-frequency curve is most similar to itself at this time delay. The position of the first target peak corresponds to the periodic characteristics of the time-frequency curve, so it can be used as the micro-motion period of the underwater target.

[0077] For example, if the position of the first target peak is 0.5 seconds, then the micro-motion period of the underwater target is 0.5 seconds, which means that the micro-motion feature of the underwater target is repeated every 0.5 seconds.

[0078] In some embodiments, the echo time-frequency diagram may be subjected to continuous wavelet transform to extract the local period of the time-frequency diagram, and the micromotion period obtained after global analysis with the autocorrelation function may be weighted and fused to determine the final micromotion period. The micromotion period result, the local period may correspond to the first period weight, the micromotion period obtained by the autocorrelation function analysis may correspond to the second period weight, the first period weight and the second period weight may be adjusted based on the noise intensity, for example, when the noise is strong, the first period weight may be increased, and the specific adjustment scale may be set according to the actual situation.

[0079] By performing autocorrelation analysis on the echo time-frequency diagram and determining the micro-motion period of the underwater target, it is easy to effectively extract the micro-motion characteristics of the target and provide important information for subsequent target recognition and classification.

[0080] Step 103, projecting the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructing an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image.

[0081] In some embodiments, in order to improve the accuracy of underwater target recognition, the time-frequency sub-images can be projected in multiple directions to capture the micro-motion features of the target from different angles, thereby obtaining more comprehensive information to improve the final target classification and recognition effect.

[0082] The projection intensity distribution can be a one-dimensional intensity distribution in each direction (such as 0°, 45°, 90°, 135°, etc.) obtained by linearly integrating or accumulating the time-frequency sub-image. Specifically, for each projection direction, all pixel values ​​in the time-frequency sub-image are accumulated along the direction to generate a one-dimensional array representing the intensity distribution in the direction.

[0083] The energy distribution image may be a two-dimensional image constructed based on projection intensity distributions in multiple different directions.

[0084] In some embodiments, the time-frequency sub-image may be first binarized to obtain a binarized time-frequency sub-image, and then the binarized time-frequency sub-image may be projected in multiple directions, thereby simplifying the image, highlighting the target area, reducing noise interference, and improving the efficiency and accuracy of subsequent processing. Exemplarily, the time-frequency sub-image may be binarized to convert each pixel in the image into black and white according to a certain threshold value, and the threshold value may be set according to actual conditions. When the pixel value is greater than the threshold value, it is set to white (such as 1 or 255); when the pixel value is less than the threshold value, it is set to black (such as 0), and so on.

[0085] Exemplarily, the multiple directions may include 0°, 45°, 90°, 135°, etc. The selected direction should cover the possible micro-motion directions of the underwater target, and the specific direction can be selected according to actual conditions.

[0086] Furthermore, the time-frequency sub-image can be subjected to Radon transformation, that is, the time-frequency sub-image is projected along multiple directions, and the time-frequency data is projected into the Radon domain, where each projection corresponds to a specific angle. Exemplarily, the formula of Radon transformation can be as follows: ; in, is the projection intensity in the Radon domain, is the time-frequency sub-image, is the distance from the origin to the line, is the direction of projection.

[0087] Further, for each projection direction, the projection intensity distribution of the Radon transform result can be calculated, and the projection intensity distribution reflects the concentration of signal energy in this direction. Exemplarily, the projection intensity distribution can be calculated by the following formula: ; in, It is the direction Projected intensity distribution on .

[0088] Furthermore, after obtaining the projection intensity distribution in all directions, the projection intensity distribution in all directions can be integrated into a two-dimensional image to form an energy distribution image. For example, a two-dimensional array can be created whose size is the same as the Radon transform result. For each direction θ, the corresponding projection intensity distribution Fill in a row or column of the energy distribution image. At the same time, color coding can be used to represent different energy levels. Usually, high energy areas are represented by bright colors, and low energy areas are represented by dark colors. After filling, the energy distribution image can be obtained.

[0089] By constructing an energy distribution image, the accuracy and reliability of underwater target identification can be significantly improved, facilitating the subsequent accurate identification of micro-Doppler features.

[0090] In some embodiments, in order to improve the accuracy and efficiency of signal processing, the linear component in the time-frequency sub-image can be identified. When the linear component has a second-order translation, the corresponding translation compensation can be performed. After correcting the frequency shift effect caused by the translation, the projection processing can be performed to avoid the real micro-Doppler feature from being covered or distorted. Exemplarily, before step 103, that is, before "projecting the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructing an energy distribution image based on the multiple projection intensity distributions", it can also include: (B.1) Perform linear transformation on the time-frequency sub-image to obtain the linear component in the time-frequency sub-image; (B.2) For each linear component, obtain the slope and intercept of each linear component, and determine the second-order translation result of the linear component based on the slope and intercept; (B.3) When the second-order translation result indicates that the linear component has a second-order translation, the frequency axis coordinates in the time-frequency sub-image are offset-adjusted based on the slope and the intercept to obtain a processed time-frequency sub-image.

[0091] Among them, the linear component can be the micro-Doppler effect caused by the translational motion (such as uniform forward and backward movement) of the underwater target, which appears as a straight line at a certain angle to the time axis in the time-frequency sub-image.

[0092] The slope may be the inclination of the straight line component in the time-frequency sub-image, that is, the rate of change of the frequency over time.

[0093] The intercept may be the intersection of the linear component and the frequency axis in the time-frequency sub-image, indicating the initial frequency of the underwater target at time t=0, that is, the intercept of the linear component is associated with the initial Doppler frequency of the underwater target.

[0094] The second-order translation result may be an analysis result of determining whether there is a second-order translation (ie, acceleration) by analyzing the linear component. If there is a second-order translation (ie, the slope changes with time), it indicates that the underwater target has not only a translation velocity but also an acceleration.

[0095] The frequency axis coordinate may be a coordinate axis representing frequency in the time-frequency sub-image.

[0096] For example, if the time-frequency sub-image appears as a straight line due to the uniform linear motion of the underwater target, Radon transform or other linear transformation methods can be used to extract the straight line component. If the slope of the identified straight line component is 2 Hz / s and the intercept is 100 Hz (i.e., 100 Hz), this indicates that the underwater target passes through the sonar equipment at a speed of 2 Hz / s (i.e., 2 Hz per second) and the initial Doppler frequency is 100 Hz.

[0097] Furthermore, if it is detected that the slope increases over time, indicating that the underwater target is accelerating away from the sonar device, the frequency axis coordinates can be adjusted by subtracting the slope from the frequency axis to obtain a processed time-frequency sub-image. For example, if the slope is 2 Hz / s, 2 Hz can be subtracted from the frequency axis to compensate for this acceleration effect.

[0098] Through the above steps, the translational component of the target can be effectively extracted and analyzed from the time-frequency sub-image, and the Doppler frequency shift caused by the translation can be compensated, thereby improving the accuracy and reliability of underwater target recognition.

[0099] Step 104, determining the target energy region with the highest cumulative projection intensity value from the energy distribution image, and obtaining a sinusoidal curve sequence corresponding to the micro-Doppler component in the time-frequency curve based on the target position of the target energy region in the energy distribution image.

[0100] In some embodiments, since the signal characteristics that are significant in multiple projection directions correspond to the micro-motion behavior of the underwater target, the micro-Doppler component in the time-frequency sub-image can be effectively identified by determining the target energy area with the highest cumulative projection intensity value, and the sinusoidal curve sequence can be extracted from the target energy area, thereby further enhancing the feature description capability and improving the accuracy of underwater target recognition.

[0101] The cumulative projection intensity value may be the cumulative intensity of the overlapping parts between projection intensity distributions in different directions in the energy distribution image. By calculating the sum of the intensities of these overlapping parts, the most significant target energy region in the energy distribution image may be found.

[0102] The target energy region may be a local region with the highest cumulative projection intensity value in the energy distribution image, and the target energy region corresponds to a main micro-Doppler feature in the time-frequency sub-image.

[0103] The target position may be the most important coordinate position of the target energy region in the energy distribution image, which may help locate the key micro-Doppler component in the time-frequency curve.

[0104] The sinusoidal curve sequence can be a sinusoidal waveform corresponding to the target micro-motion form extracted from the time-frequency curve. The sinusoidal curve sequence reflects the micro-motion characteristics of the target (such as vibration, rotation, etc.), which is a key feature for subsequent classification and recognition.

[0105] Please refer to Figure 3In some embodiments, after projecting the time-frequency sub-image in multiple directions (converting the time-frequency sub-image from the spatial domain to the Radon domain, i.e., the projection domain) to obtain the energy distribution image, the energy distribution image can be inversely Radon transformed, i.e., the data is converted back to the spatial domain, the time-frequency sub-image is reconstructed, and in the reconstructed time-frequency sub-image, the target energy region with the largest cumulative projection intensity value is identified, such as Figure 3 As shown, Figure 3 is a simplified diagram of the reconstructed time-frequency sub-image, Figure 3 The target energy region is marked in FIG. 1 , and the target energy region corresponds to the micro-Doppler component in the echo time-frequency diagram. Alternatively, the reconstructed video sub-image may also have a black background, and the target energy region is highlighted in white.

[0106] In some embodiments, since the target energy region with the largest cumulative projection intensity value in the reconstructed time-frequency sub-image appears as a brighter region, the region with the highest pixel brightness in the reconstructed time-frequency sub-image can be directly identified as the target energy region.

[0107] For example, each pixel region (that is, each pixel point) of the reconstructed time-frequency sub-image can be traversed, and the corresponding projection intensity cumulative value can be determined for each pixel region. After that, the projection intensity cumulative values ​​of all pixel regions are compared to determine the target energy region with the highest projection intensity cumulative value, and the target energy region corresponds to the characteristics of the micro-Doppler component.

[0108] In some embodiments, when the target energy region contains multiple pixel points, the centroid method can be applied to calculate the centroid coordinates of the entire target energy region as the target position when determining the target position, or the pixel point position with the highest cumulative projection intensity value in the target energy region can be further calculated as the target position, and so on.

[0109] For example, if there are multiple high-energy areas, multiple local maximum points can be selected, corresponding to different micro-Doppler components. Specifically, an energy threshold can be set, and all projection intensity cumulative values ​​are compared with the energy threshold. The high-energy area corresponding to the projection intensity cumulative value higher than the energy threshold can also be determined as the target energy area.

[0110] Furthermore, since the micro-Doppler component is in the form of a sinusoidal curve in the time-frequency sub-image, the sinusoidal curve sequence corresponding to the micro-Doppler component can be determined by the position of the target energy region. , calculate the amplitude and initial phase of the sine curve. Specifically, the amplitude The calculation formula is as follows: ; Specifically, the initial phase The calculation formula is as follows: ; Furthermore, the sine curve sequence can be determined based on the amplitude and initial phase of the sine curve. : ; Where f is the frequency of the micro-Doppler component, is the initial phase, is the amplitude and t is the time variable.

[0111] By utilizing the detailed micro-motion information contained in the overlapping areas generated by the vibration or rotation of the target in multiple directions and calculating the cumulative intensity values ​​of these areas, the target energy areas representing the characteristics of the micro-Doppler component can be quickly located and identified. This not only speeds up the calculation speed, but also can process complex high-energy areas, ensuring the accurate extraction of micro-Doppler components in the form of sinusoidal curves from time-frequency sub-images, providing reliable data support for subsequent analysis.

[0112] Step 105 : extracting a micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generating a micro-Doppler time-frequency diagram according to the micro-Doppler component.

[0113] In some embodiments, in order to provide high-quality data support for subsequent classification and recognition tasks, the micro-Doppler component can be accurately extracted from the echo time-frequency diagram based on the sinusoidal curve sequence to focus on the micro-Doppler component, reduce the interference of other irrelevant information, and improve the accuracy and robustness of underwater target recognition.

[0114] Among them, the micro-Doppler component can be the frequency change caused by the micro-motion of the local part (such as vibration, rotation, etc.) of the underwater target during its movement. These frequency changes usually appear as specific patterns in the echo time-frequency diagram, which are different from the main translational frequency (Doppler frequency) of the target. Therefore, in order to accurately identify underwater targets, the micro-Doppler component can be extracted from the echo time-frequency diagram to avoid interference from irrelevant signals.

[0115] Among them, the micro-Doppler time-frequency diagram can be the part containing only the micro-Doppler component extracted from the echo time-frequency diagram, which shows the distribution of the micro-motion characteristics of the underwater target in time and frequency, and removes the main translational frequency components and other noise interference.

[0116] In some implementations, the generated sinusoidal curve sequence can be mixed with the original time-frequency curve to highlight the sinusoidal curve. After mixing, the micro-Doppler component moved to the baseband after mixing can be effectively separated by a low-pass filter with an appropriate cutoff frequency, while suppressing other interference signals.

[0117] In some embodiments, in order to ensure that all micro-Doppler components are completely extracted, the energy of the remaining signal can be calculated after each extraction of the micro-Doppler component (i.e., the sine curve). If the remaining energy is lower than a preset threshold, it indicates that all significant micro-Doppler components have been extracted. Otherwise, it is necessary to continue to extract the micro-Doppler components. Alternatively, the rate of change of the sine curve parameters (such as frequency, amplitude, phase) in two consecutive iterations can also be calculated. If the rate of change is less than the rate of change threshold, it may indicate that the last curve has been reached and no further extraction is required. Otherwise, it is necessary to continue to extract the micro-Doppler components.

[0118] Please refer to Figure 4 In some embodiments, when it is necessary to continue to extract the micro-Doppler component, the above steps can be repeated, that is, after extracting the micro-Doppler component from the echo time-frequency map, the micro-motion period of the underwater target is repeatedly determined according to the time-frequency curve in the updated echo time-frequency map, and at least one time-frequency sub-image corresponding to the micro-motion period is determined from the echo time-frequency map, and the time-frequency sub-image is projected in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and an energy distribution image is constructed based on the multiple projection intensity distributions, and the target energy area with the highest cumulative projection intensity value is determined from the energy distribution image, and based on the target position of the target energy area in the energy distribution image, a sinusoidal curve sequence corresponding to the micro-Doppler component in the time-frequency curve is obtained, and based on the sinusoidal curve sequence, the micro-Doppler component is extracted from the echo time-frequency map, and a micro-Doppler time-frequency map is generated according to the micro-Doppler component, until the micro-Doppler component is extracted. It should be noted that the implementation details of this part have been expanded above and will not be repeated here.

[0119] Furthermore, the extracted micro-Doppler component may be mapped to the time-frequency domain to generate a micro-Doppler time-frequency diagram, in which the horizontal axis may represent time, the vertical axis represents frequency, and the color or brightness represents the strength or energy of the signal.

[0120] By extracting the micro-Doppler component and generating the micro-Doppler time-frequency diagram in the above manner, the inverse Radon transform, which is mainly used in the imaging field, can be applied to the extraction of micro-Doppler features of underwater targets and target recognition. Unlike the recognition of underwater targets through simple signal processing (such as filtering), the present application can effectively utilize the characteristics of the micro-Doppler component after projection, accurately extract the micro-Doppler component from the echo time-frequency diagram, and make the extraction of the micro-Doppler component more targeted and more accurate, which has a wide range of engineering practical value and promotes the development of underwater target detection technology.

[0121] Step 106: input the micro-Doppler time-frequency diagram into the pre-trained classification model to obtain the recognition result of the underwater target.

[0122] In some implementations, in order to accurately identify the type and micro-motion form of underwater targets, a classification model trained with a large amount of data may be used to classify and identify the micro-Doppler time-frequency diagram to improve recognition efficiency and reliability.

[0123] The classification model can be a machine learning or deep learning model for mapping the micro-Doppler time-frequency graph to a predefined category label. The classification model is trained with a large number of sample micro-Doppler time-frequency graphs of different categories and has good classification power.

[0124] The recognition result may be a predicted category label output by the classification model after processing the micro-Doppler time-frequency diagram, which indicates the micro-motion form that the component model believes the input data corresponds to.

[0125] Exemplarily, the classification model may be a ResNet-18 model, or other models in the ResNet series, or other deep learning models, and the embodiments of the present application do not impose specific limitations on this.

[0126] Please refer to Figure 5 ,Specifically, the recognition results of underwater targets can include four micro-motion forms of ,vibration, rotation, tumbling, and cone motion, or can be adjusted according to the ,actual classification model and training parameters.

[0127] The embodiment of the present application obtains the initial echo signal reflected by the underwater target in response to the sonar signal, and generates an echo time-frequency diagram based on the initial echo signal; determines the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determines at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; projects the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructs an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image; determines the target energy area with the highest cumulative projection intensity value from the energy distribution image, and obtains a sinusoidal curve sequence corresponding to the micro-Doppler component in the time-frequency curve based on the target position of the target energy area in the energy distribution image; extracts the micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generates a micro-Doppler time-frequency diagram based on the micro-Doppler component; inputs the micro-Doppler time-frequency diagram into a pre-trained classification model to obtain the recognition result of the underwater target. In this way, the micro-motion period of the underwater target can be determined from the echo time-frequency diagram to better understand the inherent motion characteristics of the underwater target, thereby improving the accuracy of recognition. In addition, the main distribution direction of the micro-Doppler feature is consistent, and it will appear more concentrated in the projected space. By determining the target position of the target energy area with the highest cumulative projection intensity in the energy distribution image, the sinusoidal curve sequence corresponding to the micro-Doppler component can be accurately determined to accurately separate the micro-Doppler component corresponding to the underwater target from the complex echo signal for identification and classification by the classification model, providing the classification model with clearer and more representative features, effectively reducing the impact of interference signals, and improving the accuracy of recognition. In summary, the present application can improve the accuracy of underwater target recognition.

[0128] In some implementations, in order to enable the classification model to more accurately predict the sample category label and improve the classification accuracy, the preset classification model can be trained multiple times to gradually reduce the target loss, and finally obtain a classification model with excellent performance and higher classification accuracy. Exemplarily, the classification model can be trained in the following ways: (C.1) Obtain multiple first echo time-frequency diagrams corresponding to underwater targets in actual environments; the first echo time-frequency diagrams correspond to multiple micro-motion forms; (C.2) performing sample expansion based on a plurality of first echo time-frequency diagrams through a preset diffusion model to obtain a plurality of corresponding second echo time-frequency diagrams; (C.3) extracting micro-Doppler components from the first echo time-frequency diagram and the second echo time-frequency diagram respectively to obtain a plurality of sample micro-Doppler time-frequency diagrams, and generating a sample training set based on the plurality of sample micro-Doppler time-frequency diagrams; (C.4) Input each sample micro-Doppler time-frequency graph in the sample training set into the preset classification model in turn to obtain the predicted category label corresponding to each sample data; (C.5) Obtain the sample category label of the sample data and determine the target loss based on the difference between the predicted category label and the sample category label; (C.6) Based on the target loss, the parameters of the preset classification model are adjusted to obtain the classification model.

[0129] The first echo time-frequency diagram may be an image obtained after time-frequency transformation of a real echo signal collected by a sonar device in an actual environment.

[0130] Among them, the diffusion model can be a time-series diffusion model, which is used to generate new samples (second echo time-frequency diagram) with the same statistical characteristics as the original sample echo signal. This can effectively solve the problem of insufficient sonar echo data and improve the accuracy of underwater target recognition.

[0131] Among them, the second echo time-frequency diagram can be simulated data generated by the diffusion model, which represents the image of the echo signal under different micro-motion forms (such as vibration, rotation, etc.) after time-frequency transformation, and is used to enhance the data volume and diversity of the training set, so that the preset model has better generalization performance.

[0132] The actual environment may be a real underwater environment.

[0133] The sample micro-Doppler component may be a micro-Doppler feature portion extracted from the first echo time-frequency diagram and the second echo time-frequency diagram, respectively. The sample micro-Doppler component reflects the frequency change generated by a local part of the corresponding underwater target (such as engine vibration, rotor blade rotation, etc.).

[0134] Among them, the sample micro-Doppler time-frequency diagram can be a partial time-frequency diagram containing only micro-Doppler components extracted from the first echo time-frequency diagram and the second echo time-frequency diagram, which shows the distribution of the micro-motion characteristics of the underwater target in time and frequency, and removes the main translational frequency components and other noise interference.

[0135] Among them, the preset classification model can be a deep learning model, such as ResNet-18, etc., which is used to classify the input sample micro-Doppler time-frequency diagram.

[0136] The predicted category label may be a category prediction result output by a preset classification model after processing the input sample micro-Doppler time-frequency diagram, and each predicted category label indicates the probability that the preset classification model considers that the input data belongs to the corresponding micro-motion category.

[0137] The sample category label may be the actual category label of each sample data in the sample training set, indicating the true category to which the sample data belongs, and is used to evaluate the performance of the model.

[0138] The target loss may be a measure of the difference between the class label predicted by the preset classification model and the class label of the sample. The target loss may be a cross entropy loss, etc., which is used to guide the adjustment of the parameters of the preset component model to minimize the error.

[0139] Please refer to Figure 6 , the process of training the classification model is introduced. Furthermore, after obtaining the sample echo signal of the underwater target, the sample echo signal can be bandpass filtered and preprocessed, and when the sample time-frequency sub-image has a second-order translation, the translation compensation is performed, and then the sample micro-Doppler component is extracted to generate a sample micro-Doppler time-frequency diagram corresponding to each sample echo signal. Furthermore, a test set, a validation set, and a training set can be generated according to multiple sample micro-Doppler time-frequency diagrams corresponding to multiple sample echo signals.

[0140] Furthermore, a preset diffusion model can be used to perform sample data enhancement based on multiple first echo time-frequency diagrams of various micro-motion forms to obtain multiple second echo time-frequency diagrams. After processing the multiple second echo time-frequency diagrams to obtain corresponding multiple sample micro-Doppler time-frequency diagrams, they can be added to a training set for data enhancement. Afterwards, the training set can be used to train a preset classification model to obtain the recognition results output by the preset classification model.

[0141] Exemplarily, multiple first echo time-frequency diagrams corresponding to the actual environment can be divided into a test set, a validation set and a training set. After multiple second echo time-frequency diagrams are generated by a trained diffusion model, the multiple second echo time-frequency diagrams are added to the training set for data enhancement, and the preset classification model is trained based on the training set.

[0142] In some embodiments, the process of extracting sample micro-Doppler components from the first echo time-frequency diagram and the second echo time-frequency diagram, and generating a sample micro-Doppler time-frequency diagram based on the sample micro-Doppler components is the same as the process of extracting micro-Doppler components from the echo time-frequency diagram and generating a micro-Doppler time-frequency diagram based on the micro-Doppler components in the above text. The embodiment of the present application does not elaborate on the processing process of the sample micro-Doppler components in the training process, and the specific details can be referred to the aforementioned processing process.

[0143] In some implementations, the target loss may be mean square error loss, cross entropy loss, square absolute error loss, etc. The form of the specific loss function may be determined based on actual conditions. Taking the target loss as cross entropy loss as an example, the target loss L may be calculated using the following formula: ; in, is the number of categories, is the actual sample category label, is the probability of the predicted category label i predicted by the preset classification model.

[0144] Furthermore, the gap between the predicted category label and the sample category label predicted by the preset classification model can be determined based on the target loss, and the parameters of the preset classification model can be adjusted based on the target loss until the preset convergence condition is reached, and a trained classification model can be obtained. The convergence condition can be that the number of training times reaches a preset number of times, such as 500 times, 1000 times, etc., or that the target loss is less than the loss threshold, and the embodiment of the present application does not impose specific restrictions on this.

[0145] By performing data enhancement on each category, the sample micro-Doppler time-frequency diagram of the second echo time-frequency diagram is used to train the preset classification model, which can effectively enhance the classification model's ability to recognize micro-Doppler features, so as to improve the accuracy and efficiency of recognition, thereby providing important support for fields such as marine science, safety monitoring and resource management.

[0146] In some embodiments, in order to solve the problem of being unable to effectively classify the micro-motion forms of underwater targets, the present application uses a time series diffusion model to enhance the data of samples by considering the time series characteristics of echoes of different basic micro-motion forms, so that the demand for the number of original samples is greatly reduced, and the recognition accuracy is effectively improved. Exemplarily, (C.2) may include: (C.1.1) For each first echo time-frequency diagram, convert the first echo time-frequency diagram into a one-dimensional time series; (C.1.2) Inputting the one-dimensional time series into a preset diffusion model, so that the diffusion model is expanded based on the first echo time-frequency diagram to obtain multiple second echo time-frequency diagrams with the same micro-motion form as the first echo time-frequency diagram; wherein the first dimension of each convolution kernel in the diffusion model is adapted to the dimension of the one-dimensional time series.

[0147] The first echo signal may be original sonar echo data acquired from an underwater target.

[0148] The first echo time-frequency diagram may be a two-dimensional image obtained by performing time-frequency transformation (such as short-time Fourier transform or wavelet transform) on the first echo signal. Different first echo signals contain various micro-motion forms of underwater targets (such as vibration, rotation, tumbling and cone motion) and environmental noise information.

[0149] The one-dimensional time series may be a form of converting the first echo time-frequency diagram into a one-dimensional array. The two-dimensional first echo time-frequency diagram may be compressed into a one-dimensional array by projecting the first echo time-frequency diagram along a specific direction or processing it in other ways.

[0150] The first dimension of the convolution kernel can be the size of the first dimension of the input data used to process the convolution layer of the diffusion model. For a one-dimensional time series input, the first dimension of the convolution kernel should be adapted to the length of the time series.

[0151] Please refer to Figure 7 . For example, Figure 7 is the structural diagram of the diffusion model. Specifically, The input data for the diffusion model can be a raw, unprocessed one-dimensional time series, from Initially, the diffusion model can gradually add noise during the forward diffusion process. to the data to generate a series of noisy data , to gradually change the data distribution from the original distribution to the Gaussian noise distribution.

[0152] Furthermore, at each time step t, a neural network (U-net) can be used to predict the noise added at that step , U-net learns to remove noise from partially noisy data, thereby generating data that is closer to the original distribution.

[0153] Furthermore, from the noisy data Start by gradually removing the noise and generating a series of data that gradually approaches the original distribution , in order to simulate the process of gradually recovering from the Gaussian noise distribution to the original data distribution. Afterwards, at each time step t, the loss function is calculated To measure the noise of the model prediction The actual noise added By minimizing these losses, the U-net learns to more accurately predict the noise added at each step. Ultimately, the data generated by the diffusion model It should be a one-dimensional time series with the original input New data that is similar to or conforms to the one-dimensional time series distribution. In this way, the diffusion model can learn to generate high-quality new samples from noisy data, thereby solving the problem of data scarcity and improving the accuracy of underwater target recognition.

[0154] In some embodiments, in order to enhance and generate data using a diffusion model, the first echo time-frequency diagram may be converted into a one-dimensional time series so that the time-frequency data fits the diffusion model, thereby reducing the time and resource consumption of diffusion model reasoning.

[0155] In some embodiments, each pixel value of the first echo time-frequency diagram may be arranged in row or column order to form a one-dimensional time series, so that the diffusion model can perform time series analysis to capture the time dependency of the echo signal by simulating the evolution of data over time.

[0156] In some embodiments, since the diffusion model is generally used to process two-dimensional or three-dimensional image data, its convolution kernel is designed to be two-dimensional or three-dimensional. Therefore, in the process of generating samples through the diffusion model, the present application needs to no longer apply the first dimension (spatial dimension) of the convolution kernel. In order to adapt to the processing of one-dimensional time series, the first dimension of the convolution kernel is set to 1 to match the characteristics of the time series data, so that the diffusion model can apply the convolution operation on the time series and capture the timing characteristics in the sequence (here, it is considered that the sonar echo data has time dependence, that is, the value of the current time point is usually related to the value of the previous one or more time points), and generate a more accurate second echo time-frequency diagram.

[0157] In some embodiments, in order to enable the classification model to fully capture the complex timing characteristics and dynamic behaviors in the sample, the first echo time-frequency diagram corresponding to various micro-motion forms can be diffused to obtain a large number of second echo time-frequency diagrams. After the preset classification model is trained with multiple second echo time-frequency diagrams, the classification model can better adapt to complex environments and improve the performance of the classification model in real underwater environments.

[0158] By applying the diffusion model to enhance the data set for underwater target recognition based on the first echo time-frequency diagram in various micro-motion forms and performing time series processing on the first echo time-frequency diagram, the diffusion model can focus on time series feature extraction and generate rich samples, effectively solving the problem of small number and difficulty in obtaining underwater samples, improving the recognition performance and accuracy of the model, and optimizing the use of computing resources, making it easier to improve the robustness of the classification model through a large number of samples in the future.

[0159] See also Figure 8 The embodiment of the present application further provides an underwater target recognition device, which can implement the above-mentioned underwater target recognition method. The underwater target recognition device includes: An acquisition module 81 is used to acquire an initial echo signal reflected by an underwater target in response to a sonar signal, and generate an echo time-frequency diagram based on the initial echo signal; A first determination module 82 is used to determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; A construction module 83 is used to project the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and to construct an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image; A second determination module 84 is used to determine the target energy region with the highest cumulative projection intensity value from the energy distribution image, and obtain a sinusoidal curve sequence corresponding to the micro-Doppler component in the time-frequency curve based on the target position of the target energy region in the energy distribution image; A generating module 85, configured to extract a micro-Doppler component from the echo time-frequency diagram based on a sinusoidal curve sequence, and generate a micro-Doppler time-frequency diagram according to the micro-Doppler component; The input module 86 is used to input the micro-Doppler time-frequency diagram into the pre-trained classification model to obtain the recognition result of the underwater target.

[0160] The specific implementation of the underwater target recognition device is basically the same as the specific implementation of the underwater target recognition method described above, and will not be repeated here. Under the premise of meeting the requirements of the embodiment of the present application, the underwater target recognition device can also be provided with other functional modules to implement the underwater target recognition method in the above embodiment.

[0161] The embodiment of the present application also provides a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the underwater target recognition method when executing the computer program. The computer device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0162] See also Fig. 9 , Fig. 9 The hardware structure of a computer device according to another embodiment is shown, and the computer device includes: The processor 91 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application; The memory 92 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 92 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 92, and the processor 91 calls and executes the underwater target recognition method of the embodiment of the present application; Input / output interface 93, used to implement information input and output; Communication interface 94, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.); A bus 95 that transmits information between the various components of the device (e.g., the processor 91, the memory 92, the input / output interface 93, and the communication interface 94); The processor 91 , the memory 92 , the input / output interface 93 and the communication interface 94 are connected to each other in communication within the device via a bus 95 .

[0163] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned underwater target recognition method is implemented.

[0164] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0165] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0166] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0167] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0168] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0169] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0170] It should be understood that in the present application, "at least one (item)" and "several" refer to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0171] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0172] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0173] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0174] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0175] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for underwater target recognition, characterized in that: The method comprises: Acquire an initial echo signal reflected by an underwater target in response to a sonar signal, and generate an echo time-frequency diagram based on the initial echo signal; Determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; Projecting the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and constructing an energy distribution image based on the multiple projection intensity distributions; Wherein, there are overlapping projections between different projection intensity distributions in the energy distribution image; Determine a target energy region having the highest cumulative projection intensity value from the energy distribution image, and obtain a sinusoidal curve sequence corresponding to a micro-Doppler component in the time-frequency curve based on a target position of the target energy region in the energy distribution image; Extracting a micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generating a micro-Doppler time-frequency diagram according to the micro-Doppler component; The micro-Doppler time-frequency diagram is input into a pre-trained classification model to obtain a recognition result of the underwater target.

2. The underwater target recognition method according to claim 1, characterized in that: The step of generating an echo time-frequency diagram based on the initial echo signal comprises: Acquiring a radial displacement velocity of the underwater target, a sound velocity of an underwater environment corresponding to the underwater target, and a carrier frequency of the sonar signal, and determining a predicted Doppler frequency of the underwater target based on the radial displacement velocity, the sound velocity, and the carrier frequency; Obtaining a preset filter range coefficient, and setting upper and lower cutoff frequencies of a bandpass filter based on the carrier frequency, the filter range coefficient and the predicted Doppler frequency; The initial echo signal is filtered using the upper and lower cutoff frequencies to obtain a first echo signal; performing normalization processing on the first echo signal to obtain a second echo signal; Perform time-frequency transformation on the second echo signal to obtain an echo time-frequency diagram corresponding to the initial echo signal.

3. The underwater target recognition method according to claim 2, characterized in that: Before the first echo signal is normalized to obtain the second echo signal, the method further includes: Sequentially intercepting signals to be processed from the first echo signal through a preset sliding window; For the signal to be processed corresponding to each sliding window, obtain the mean and standard deviation of the signal to be processed, and set a threshold range of the signal to be processed based on the mean and the standard deviation; When a target signal value in the signal to be processed is outside the threshold range, replacing the target signal value with the mean value; After the processing of the multiple signals to be processed contained in the first echo signal is completed, a processed first echo signal is obtained.

4. The underwater target recognition method according to claim 1, characterized in that: The step of determining the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram comprises: Performing autocorrelation analysis on the echo time-frequency diagram to obtain an autocorrelation function corresponding to the initial echo signal; wherein the autocorrelation function is used to measure the degree of overlap of the time-frequency curves under different time delays; Obtaining a first target peak value of the autocorrelation function; Based on the position of the first target peak, the micro-motion period of the underwater target is determined.

5. The underwater target recognition method according to claim 1, characterized in that: Before projecting the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions and constructing an energy distribution image based on the multiple projection intensity distributions, the method further includes: Performing a linear transformation on the time-frequency sub-image to obtain a linear component in the time-frequency sub-image; For each straight line component, obtain the slope and intercept of each straight line component, and determine the second-order translation result of the straight line component based on the slope and the intercept; When the second-order translation result indicates that the linear component has a second-order translation, the frequency axis coordinates in the time-frequency sub-image are offset-adjusted based on the slope and the intercept to obtain a processed time-frequency sub-image.

6. The underwater target recognition method according to claim 1, characterized in that: The classification model is trained in the following way: Acquire multiple first echo time-frequency diagrams corresponding to the underwater target in the actual environment; the first echo time-frequency diagrams correspond to multiple micro-motion forms; By using a preset diffusion model, sample expansion is performed based on a plurality of first echo time-frequency diagrams to obtain a plurality of corresponding second echo time-frequency diagrams; Extracting micro-Doppler components from the first echo time-frequency diagram and the second echo time-frequency diagram respectively to obtain a plurality of sample micro-Doppler time-frequency diagrams, and generating a sample training set based on the plurality of sample micro-Doppler time-frequency diagrams; Inputting each sample micro-Doppler time-frequency graph in the sample training set into a preset classification model in turn to obtain a predicted category label corresponding to each sample data; Obtaining a sample category label of the sample data, and determining a target loss based on a difference between the predicted category label and the sample category label; Based on the target loss, the parameters of the preset classification model are adjusted to obtain a classification model.

7. The underwater target recognition method according to claim 6, characterized in that: The method of performing sample expansion based on a plurality of first echo time-frequency diagrams through a preset diffusion model to obtain a plurality of corresponding second echo time-frequency diagrams includes: For each first echo time-frequency diagram, converting the first echo time-frequency diagram into a one-dimensional time series; The one-dimensional time series is input into a preset diffusion model so that the diffusion model is expanded based on the first echo time-frequency diagram to obtain a plurality of second echo time-frequency diagrams having the same micro-motion form as the first echo time-frequency diagram; wherein the first dimension of each convolution kernel in the diffusion model is adapted to the dimension of the one-dimensional time series.

8. An underwater target recognition device, characterized in that: The device comprises: An acquisition module, used for acquiring an initial echo signal reflected by an underwater target in response to a sonar signal, and generating an echo time-frequency diagram based on the initial echo signal; A first determination module is used to determine the micro-motion period of the underwater target according to the time-frequency curve in the echo time-frequency diagram, and determine at least one time-frequency sub-image corresponding to the micro-motion period from the echo time-frequency diagram; A construction module, used to project the time-frequency sub-image in multiple directions to obtain multiple projection intensity distributions of the time-frequency sub-image in different projection directions, and construct an energy distribution image based on the multiple projection intensity distributions; wherein there are overlapping projections between different projection intensity distributions in the energy distribution image; A second determination module is used to determine a target energy region with the highest cumulative projection intensity value from the energy distribution image, and obtain a sinusoidal curve sequence corresponding to a micro-Doppler component in the time-frequency curve based on a target position of the target energy region in the energy distribution image; A generating module, configured to extract a micro-Doppler component from the echo time-frequency diagram based on the sinusoidal curve sequence, and generate a micro-Doppler time-frequency diagram according to the micro-Doppler component; An input module is used to input the micro-Doppler time-frequency diagram into a pre-trained classification model to obtain a recognition result of the underwater target.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the underwater target recognition method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the underwater target recognition method according to any one of claims 1 to 7 is implemented.

Images