A signal feature extraction method and device, a terminal device, and a storage medium

Abstract

This application provides a signal feature extraction method, apparatus, terminal device, and storage medium, applicable to the field of acoustic signal processing technology. The method includes: acquiring an underwater acoustic signal of a target to be detected; extracting temporal features from the underwater acoustic signal based on a preset time step, preset state variables, a preset gate weight matrix, a preset gate bias variable, and a preset normalization function to generate initial signal features; enhancing the initial signal features based on multiple preset linear transformation matrices to generate intermediate signal features; and classifying and calculating the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function to generate target signal features. This application automatically performs preliminary feature extraction on the underwater target acoustic signal based on the temporal correlation of the acoustic signal, and filters out noise in the signal features through enhancement and classification processing, thereby improving the comprehensiveness and accuracy of signal feature extraction.

Description

Technical Field

[0001] This application belongs to the field of acoustic signal processing technology, and in particular relates to signal feature extraction methods, devices, terminal equipment and storage media. Background Technology

[0002] The demand for underwater resource exploration, development, and maintenance is growing daily. Underwater target detection using acoustic technology, as a crucial component of underwater operations, has a promising future. For underwater target detection, the extraction of underwater target acoustic signal features is indispensable. However, the complex and variable aquatic environment, including low visibility, frequent biological activity, and extreme temperature and pressure variations, presents significant challenges to underwater target signal feature extraction.

[0003] In existing technologies, acoustic signals are typically feature extracted based on the time domain or frequency domain. For example, the acoustic signal is divided into multiple short frames, and each frame is processed by pre-emphasis, Fourier transform, Mel filtering, etc., and then the Mel spectrum cepstral coefficients are calculated. Alternatively, the acoustic signal is visualized directly in the time domain, and features such as short-time energy, short-time zero-crossing rate, or autocorrelation are obtained from the time domain waveform.

[0004] However, existing methods for extracting sound signals typically do not consider the correlation between sound signals and are highly sensitive to noise, making it difficult to extract sound signals comprehensively and effectively in complex aquatic environments. Summary of the Invention

[0005] In view of this, embodiments of this application provide a signal feature extraction method, apparatus, terminal device, and storage medium, aiming to solve the problems in the prior art that ignore the correlation between sound signals, the extraction process cannot eliminate noise, and thus the accuracy and effectiveness of sound signal feature extraction results are low.

[0006] The first aspect of this application provides a signal feature extraction method, including:

[0007] Acquire underwater acoustic signals from the target to be detected;

[0008] Based on multiple preset time step information, preset state variables, preset gate weight matrix, preset gate bias variable, and preset normalization function, the underwater sound signal is subjected to time feature extraction to generate initial signal features.

[0009] The initial signal features are enhanced based on multiple preset linear transformation matrices to generate intermediate signal features.

[0010] Based on multiple randomly generated set center points and a preset set spatial displacement function, the intermediate signal features are classified and calculated to generate target signal features.

[0011] A second aspect of this application provides a signal feature extraction apparatus, comprising:

[0012] The underwater sound signal acquisition module is used to acquire the underwater sound signal of the target to be detected.

[0013] The initial signal feature generation module is used to extract time-series features from the underwater sound signal based on multiple preset time-series step-size information, preset state variables, preset gate weight matrix, preset gate bias variables, and preset normalization function, and generate initial signal features.

[0014] An intermediate signal feature generation module is used to enhance the initial signal features based on multiple preset linear transformation matrices to generate intermediate signal features; and

[0015] The target signal feature generation module is used to classify and calculate the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function, and generate target signal features.

[0016] A third aspect of this application provides a terminal device, the terminal device including a memory and a processor, the memory storing a computer program executable on the processor, and the processor executing the computer program to implement the steps of the signal feature extraction method described in the first aspect above.

[0017] A fourth aspect of this application provides a computer-readable storage medium, comprising: storing a computer program, wherein when executed by a processor, the computer program implements the steps of the signal feature extraction method described in the first aspect above.

[0018] The beneficial effects of this application embodiment compared with the prior art are as follows: considering the autocorrelation and cross-correlation of each sound signal sequence, preliminary feature extraction is performed on the underwater sound information of the target to be detected, while capturing the time domain, frequency domain and temporal features of the sound signal. Then, the preliminary extracted features are subjected to linear temporal feature enhancement processing and classification processing to filter out noise and redundant information in the signal features, thereby improving the comprehensiveness, effectiveness and accuracy of signal feature extraction. This enables comprehensive and effective extraction of sound signals in complex aquatic environments, thereby improving the accuracy and robustness of underwater target detection through the extracted sound signal features. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 1 of this application;

[0021] Figure 2 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 2 of this application;

[0022] Figure 3 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 3 of this application;

[0023] Figure 4 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 4 of this application;

[0024] Figure 5 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 5 of this application;

[0025] Figure 6 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment Six of this application;

[0026] Figure 7 This is a schematic diagram illustrating the implementation process of the signal feature extraction method provided in Embodiment 7 of this application;

[0027] Figure 8 This is a schematic diagram of the signal feature extraction device provided in the embodiments of this application;

[0028] Figure 9 This is a schematic diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0029] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0030] To illustrate the technical solution described in this application, specific embodiments are provided below.

[0031] Figure 1A flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 1 of this application is shown, and is described in detail below:

[0032] Step S101: Acquire the underwater acoustic signal of the target to be detected.

[0033] In this embodiment, underwater sound signals can be collected by a sonar device placed in the underwater environment. The underwater sound information is obtained by the sonar device emitting sound waves and receiving the reflected signals of the target to be detected. The sonar device can send the reflected signals to the server, thereby realizing the acquisition of the underwater sound signals of the target to be detected.

[0034] Step S102: Based on multiple preset time step length information, preset state variables, preset gate weight matrix, preset gate bias variables, and preset normalization function, the underwater sound signal is subjected to time sequence feature extraction to generate initial signal features.

[0035] In this embodiment, the preset time step information, preset state variables, preset gate weight matrix, and preset gate bias variables can all be manually set. The preset state variables, preset gate weight matrix, and preset gate bias variables can also be obtained by training a specific computational model and then extracting the parameters from the trained model. The preset standardization function can be designed based on an exponential function to standardize the calculation results to values ​​within the interval (0, 1), preventing invalid values ​​due to values ​​exceeding the server's computational scale. The underwater sound signal can be segmented and extracted using the time step information. The extracted data segments are then format-converted into multiple data matrices. These data matrices are then transformed multiple times using the state variables, gate weight matrix, and gate bias variables. The calculation results are used as the independent variables of the standardization function, and the calculated function value is used as the initial signal feature, thus achieving the initial extraction of underwater sound signal features.

[0036] Step S103: Based on multiple preset linear transformation matrices, the initial signal features are enhanced to generate intermediate signal features.

[0037] In this embodiment, the preset linear transformation matrix can be manually set. Multiple preset linear transformation matrices can be used to transform the initial signal features in different dimensions, thereby increasing the logical distance between multiple values ​​of the initial signal features and amplifying the differences between the initial signal feature values ​​to achieve feature enhancement processing. The enhanced features serve as intermediate signal features, which are convenient for further screening to remove noise and redundant components from the extracted sound signal.

[0038] Step S104: Based on multiple randomly generated set center points and a preset set spatial displacement function, classify and calculate the intermediate signal features to generate target signal features.

[0039] In this embodiment, the preset set space displacement function can be manually set, or it can be a Griewank function, a Rastrigin function, a Schaffer function, an Ackley function, or a Rosenbrock function. The value and number of set center points can be randomly generated and used to classify intermediate signal features. Understandably, the number of sets of center points determines the number of classes the intermediate signal features will be divided into. The classification can be achieved by calculating the logical distance between each set center point and each value in the intermediate signal feature. This logical distance initially divides the values ​​in the intermediate signal feature into different sets, achieving preliminary classification. Then, the set space displacement function is used to further calculate the logical distance, adjusting the sets to which each value in the intermediate signal feature belongs, thereby achieving deep classification calculation. This is used to distinguish the effective feature values, noise component values, and redundant component values ​​in the intermediate signal feature, and outputs the set to which the effective feature values ​​belong as the target signal feature for identification and detection of targets in the underwater environment.

[0040] The signal feature extraction method provided in this application embodiment considers the autocorrelation and cross-correlation of each sound signal sequence. It performs preliminary feature extraction on the underwater sound information of the target to be detected, and simultaneously captures the time domain, frequency domain, and temporal features of the sound signal. Then, it performs linear temporal feature enhancement processing and classification processing on the preliminary extracted features, filters out noise and redundant information in the signal features, and improves the comprehensiveness, effectiveness, and accuracy of signal feature extraction. This enables comprehensive and effective extraction of sound signals in complex aquatic environments, thereby improving the accuracy and robustness of underwater target detection based on the extracted sound signal features.

[0041] Figure 2 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 2 of this application is shown. The difference between this method and Embodiment 1 described above is that:

[0042] The preset state variables include preset hidden state variables and preset updated state variables;

[0043] The preset gating weight matrix includes a preset pre-gating weight matrix, a preset first intermediate gating weight matrix, a preset second intermediate gating weight matrix, and a preset post-gating weight matrix;

[0044] The preset gating bias variables include preset pre-gating bias variables, preset first intermediate gating bias variables, preset second intermediate gating bias variables, and preset post-gating bias variables.

[0045] Step S102 specifically includes:

[0046] Step S201: Based on multiple preset timing step information, the underwater sound signal is segmented and spliced ​​to obtain multiple signal timing segment information.

[0047] In this embodiment, the preset timing step information can be set manually and used to perform segmentation operations on the numerical sequence composed of underwater sound signals at different intervals, and to splice the segmented data segments to obtain multiple signal timing segment information.

[0048] Step S202: Based on the preset pre-gating weight matrix, preset hidden state variables, preset pre-gating bias variables, and preset standardization function, the multiple signal time sequence information is mapped, transformed, and standardized to obtain the pre-gating signal time sequence feature values.

[0049] In this embodiment, the preset pre-gating weight matrix, the preset hidden state variables, and the preset pre-gating bias variables can all be pre-set manually. The preset standardization function can be a hyperbolic tangent function or a logistic function. It can be that the hidden state variables are first combined with signal timing segment information to generate a specific matrix, then this matrix is ​​multiplied by the pre-gating weight matrix, the multiplication result is added to the pre-gating bias variables, and the sum is used as the independent variable of the standardization function. The calculated value of the standardization function is then output as the pre-gating signal timing feature value.

[0050] Step S203: Based on the preset first intermediate gating weight matrix, preset hidden state variables, preset first intermediate gating bias variables, and preset standardization function, the multiple signal time sequence information is mapped, transformed, and standardized to obtain the first intermediate gating signal time sequence feature value.

[0051] In this embodiment, both the preset first intermediate gating weight matrix and the preset first intermediate gating bias variable can be manually set. Alternatively, a specific matrix can be generated by combining the hidden state variables with signal timing segment information. This matrix is ​​then multiplied by the first intermediate gating weight matrix, and the result is added to the first intermediate gating bias variable. The result is then used as the independent variable of a standardization function, and the calculated value of the standardization function is output as the first intermediate gating signal timing feature value.

[0052] Step S204: Based on the preset second intermediate gating weight matrix, the preset hidden state function, and the preset second intermediate gating bias variable, the signal timing segment information is mapped, transformed, and standardized to obtain intermediate gating signal timing feature variables.

[0053] In this embodiment, both the preset second intermediate gating weight matrix and the preset second intermediate gating bias variable can be manually set. Alternatively, a specific matrix can be generated by combining the hidden state variables with signal timing segment information. This matrix is ​​then multiplied by the second intermediate gating weight matrix, and the result is added to the second intermediate gating bias variable. This added result is then used as the independent variable of a standardization function, and the calculated value of the standardization function is output as the second intermediate gating signal timing feature value.

[0054] Step S205: Perform hyperbolic tangent transform on the timing characteristic variable of the intermediate gate signal to obtain the timing characteristic value of the second intermediate gate signal.

[0055] In this embodiment, the timing characteristic variable of the intermediate gate signal can be used as the independent variable of the hyperbolic tangent function, and the calculated function value can be output as the timing characteristic value of the second intermediate gate signal.

[0056] Step S206: Based on the preset post-gating weight matrix, preset hidden state variables, preset post-gating bias variables, and preset standardization function, the multiple signal timing segment information is mapped, transformed, and standardized to obtain the post-gating signal timing feature values.

[0057] In this embodiment, both the preset post-gating weight matrix and the preset post-gating bias variable can be manually set. Alternatively, a specific matrix can be generated by combining the hidden state variables with signal timing segment information. This matrix is ​​then multiplied by the post-gating weight matrix, and the result is added to the post-gating bias variable. The sum is then used as the independent variable of a standardized function, and the calculated value of the standardized function is output as the post-gating signal timing feature value.

[0058] Step S207: Generate initial signal features based on the timing feature values ​​of the pre-gating signal, the first intermediate gating signal, the second intermediate gating signal, and the post-gating signal.

[0059] In this embodiment, the timing feature value of the pre-gated signal can be multiplied by the timing feature value of the first intermediate gated signal, and the timing feature value of the post-gated signal can be multiplied by the timing feature value of the second intermediate gated signal, and the two multiplication results can be summed to obtain the initial signal feature.

[0060] The signal feature extraction method provided in this application embodiment is based on multiple preset time step information and uses multiple preset gate weight matrices to extract features of underwater sound signals in different dimensions. This fully extracts the time-series features of underwater sound signals that have nonlinear variation patterns in different dimensions, which facilitates the analysis of long-range relationships of underwater sound signals and improves the accuracy and robustness of underwater target detection.

[0061] Figure 3 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 3 of this application is shown. The difference between this method and Embodiment 2 is that step S207 specifically includes:

[0062] Step S301: Calculate the state representation timing feature variable based on the timing feature value of the pre-gated gate signal, the timing feature value of the first intermediate gate signal, the timing feature value of the second intermediate gate signal, and the preset update state variable.

[0063] In this embodiment, the preset update state variable can be manually set. It can be achieved by first multiplying the timing feature value of the preceding gating signal with the timing feature value of the first intermediate gating signal, then multiplying the timing feature value of the second intermediate gating signal with the update state variable, and finally summing the results of the two multiplications. The summation result is then output as the state representation timing feature variable.

[0064] Step S302: Perform hyperbolic tangent transform on the state representation time-series characteristic variables to obtain state transformation time-series characteristic variables.

[0065] In this embodiment, the state-representation time-series characteristic variable can be used as the independent variable of the hyperbolic tangent function, and the calculated function value can be used as the output of the state-transformation time-series characteristic variable. This achieves the hyperbolic tangent transformation of the state-transformation time-series characteristic variable, thereby restricting the state-transformation time-series characteristic variable to between 0 and 1. While preserving data information, this also achieves the normalization of the data, avoiding the problem of numerical invalidation due to exceeding the dimensions in subsequent calculations.

[0066] Step S303: Calculate multiple audio signal timing feature representation variables based on the timing feature values ​​of the post-gated gate signal and the timing feature variables of the state transition.

[0067] In this embodiment, the timing feature value of the post-gated signal and the timing feature variable of the state transition can be multiplied, and the result of the multiplication can be output as the timing feature representation variable of the audio signal. It is understood that both the timing feature value of the post-gated signal and the timing feature variable of the state transition can be presented in matrix form, so the multiplication calculation can be an inner product or a dot product.

[0068] Step S304: Generate initial signal features based on multiple audio signal timing feature representation variables.

[0069] In this embodiment, it can be understood that the temporal feature representation variables of the sound signal are presented in the form of multiple matrices. The matrices need to be formatted and converted into a sequence of multiple discrete values ​​before they can be used as the initial signal features for output.

[0070] The signal feature extraction method provided in this application combines and transforms the timing feature values ​​of the pre-gated signal, the timing feature values ​​of the first intermediate gated signal, the timing feature values ​​of the second intermediate gated signal, and preset update state variables. This effectively captures complex features and long-range relationships in the sound signal, enhances the feature representation capability of the sound signal, and avoids numerical invalidation due to the numerical values ​​exceeding the numerical dimensions processed by the computer during the calculation process. This facilitates further enhancement and screening of the sound feature signals, thereby improving the accuracy and effectiveness of sound signal extraction.

[0071] Figure 4 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 4 of this application is shown. Its difference from Embodiment 1 described above lies in:

[0072] The preset linear transformation matrices include a preset first linear layer transformation matrix, a preset second linear layer transformation matrix, and a preset third linear layer transformation matrix;

[0073] Step S103 specifically includes:

[0074] Step S401: Calculate the first linear layer feature representation variable, the second linear layer feature representation variable, and the third linear layer feature representation variable based on the initial signal features, the preset first linear layer transformation matrix, the preset second linear layer transformation matrix, and the preset third linear layer transformation matrix.

[0075] In this embodiment, the preset first linear layer transformation matrix, the preset second linear layer transformation matrix, and the preset third linear layer transformation matrix can all be manually set. They are used to perform multiplication operations with the initial signal features, respectively, to widen the differences between the values ​​of the initial signal features from multiple different dimensions, facilitating enhancement processing. Specifically, the initial signal features can be multiplied or convolved with the first linear layer transformation matrix, the second linear layer transformation matrix, and the third linear layer transformation matrix, respectively. The calculation results are output as the first linear layer feature representation variables, the second linear layer feature representation variables, and the third linear layer feature representation variables.

[0076] Step S402: Obtain intermediate linear representation feature variables based on the first linear layer feature representation variables and the second linear layer feature representation variables.

[0077] In this embodiment, the first linear layer feature representation variable and the second linear layer feature representation variable can be subjected to a dot product operation, and the result can be used as an intermediate linear representation feature variable to achieve a linear transformation of the sound signal features and increase the logical distance between the values ​​of each signal feature at the linear level.

[0078] Step S403: Multiply the intermediate linear representation feature variable and the third linear layer feature representation variable to obtain the signal feature linear representation variable.

[0079] In this embodiment, the intermediate linear representation feature variable and the third linear layer feature representation variable are multiplied together, and the result of the multiplication is output as the signal feature linear representation variable.

[0080] Step S404: Based on the preset linear layer weight matrix, the linear representation variables of the signal features are weighted and summed to generate intermediate signal features.

[0081] In this embodiment, the preset linear layer weight matrix can be manually set. The values ​​in the preset linear layer weight matrix are used as weights to perform a weighted summation calculation on the values ​​in the linear representation variables of the signal features, and the result of the weighted summation is used as the intermediate signal feature output.

[0082] The signal feature extraction method provided in this application performs multi-dimensional linear transformations on the initial signal features using a preset first linear layer transformation matrix, a second linear layer transformation matrix, and a third linear layer transformation matrix. This increases the logical distance between the values ​​in the initial signal features, making the contrast of the initially extracted sound signal features more obvious. This achieves enhanced processing of the initial signal features, effectively enhancing the feature representation capability of the sound signal features, and facilitating further filtering of noise and redundant components in the sound signal features.

[0083] Figure 5 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 5 of this application is shown. The difference between this method and Embodiment 1 is that step S104 specifically includes:

[0084] Step S501: Calculate the feature set representation distance between the intermediate signal feature and the feature set representation distance of the multiple set center points.

[0085] In this embodiment, the feature set representation distance can be Euclidean distance. By quantifying the logical distance between the intermediate signal feature and multiple set center points, it is convenient to subsequently divide the various numerical points of the intermediate signal feature into sets.

[0086] Step S502: Based on the feature set representation distance, the intermediate signal features are divided into multiple sets corresponding to multiple set center points to generate multiple signal feature sets.

[0087] In this embodiment, a logical distance interval can be set. When the feature set representation distance falls into the logical distance interval, the intermediate signal feature corresponding to the feature set representation distance is assigned to the set to which the set center point corresponding to the feature set representation distance belongs, thereby generating multiple signal feature sets to achieve preliminary classification of signal features.

[0088] Step S503: Calculate the spatial displacement of each intermediate signal feature based on the signal feature set and the preset spatial displacement function.

[0089] In this embodiment, the preset set space displacement function can be manually set, or it can be a Griewank function, a Rastrigin function, a Schaffer function, an Ackley function, or a Rosenbrock function. Alternatively, the set of signal features can be used as the independent variable of the set space displacement function, and the calculated set space displacement function can be used as the set space displacement amount.

[0090] Step S504: Determine the displacement center point of each of the signal feature sets based on the spatial displacement of the sets.

[0091] In this embodiment, the value with the largest spatial displacement in the signal feature set can be used as the displacement center point of the signal feature set. This can be used to reclassify intermediate signal features so that the classification result approaches the global optimal solution, and can also be used to determine whether the classification calculation has converged in the subsequent process.

[0092] Step S505: Determine whether the displacement center point is the same as the set center point; if yes, proceed to step S506; if no, proceed to step S507.

[0093] In this embodiment, when the displacement center point is the same as the set center point, it indicates that the current classification calculation process for the intermediate signal features has converged. Further division of the set to which the intermediate signal features belong is unnecessary, and therefore, no further calculation of the displacement center point is required. At this point, the output can be directly based on the set division result obtained from the set center point, and the signal feature set with at least a specific number of elements can be used as the target signal feature. When the displacement center point is different from the set center point, it indicates that the current classification calculation for the intermediate signal features has not yet converged, and further division of the set to which the intermediate signal features belong is required. Therefore, further calculation of the displacement center point is needed to make the calculation result tend towards the global optimal solution. It is understandable that intermediate signal features include noise and redundant components from the sound signal features. Noise and redundant components are random and have diverse sources, making it impossible to classify various noise and redundant components into one or several categories. Therefore, the values ​​constituting the noise and redundant components will have discretized characteristics and cannot form sets with many elements. However, the sound signal belonging to the target to be detected has a certain linear pattern and can be divided into one or several sets, thus forming multiple sets with many elements.

[0094] Step S506: Generate target signal features based on the multiple sets of signal features.

[0095] In this embodiment, when the displacement center point is the same as the set center point, it indicates that the current classification calculation process for the intermediate signal features has converged. There is no need to further divide the set to which the intermediate signal features belong. Therefore, there is no need to further calculate the displacement center point. At this time, the set input result obtained based on the set center point division can be directly output. The signal feature set with no less than a certain number of set elements can be used as the target signal feature for identification and processing of the target to be detected in the underwater environment.

[0096] Step S507: Take the displacement center point as the set center point and return to step S502.

[0097] In this embodiment, if the displacement center point is different from the set center point, it indicates that the classification calculation of the intermediate signal features has not yet converged. At present, the effective signal features have not been distinguished from noise components and redundant components in each set. Further division of the set to which the intermediate signal features belong is required, which requires further calculation of the displacement center point so that the calculation result tends to the global optimal solution.

[0098] The signal feature extraction method provided in this application performs depth classification calculation on intermediate signal features by randomly generating set center points and preset set spatial displacement functions. It effectively distinguishes between effective features, noise components, and redundant components in the intermediate signal features, thereby selecting effective components in the intermediate signal features for underwater target detection. This achieves the filtering out of noise and redundant components, improves the effectiveness and accuracy of sound signal feature extraction, and enhances the accuracy and robustness of target detection in complex and ever-changing underwater environments.

[0099] Figure 6 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment Six of this application is shown. The difference between this method and Embodiment One described above is that, after step S104, the method further includes:

[0100] Step S601: Extract the time step length information and the number of set center points corresponding to multiple target signal features.

[0101] In this embodiment, it is understood that there are multiple time step lengths used to extract the temporal features of the sound signal, and multiple set center points used to classify the intermediate signal features. All possible values ​​for the number of time step lengths and set center points are extracted for subsequent screening of the optimal configuration values ​​for the number of time step lengths and set center points, in order to select the combination of time step lengths and set center points that yields the best quality of the extracted signal features.

[0102] Step S602: Generate an initial population based on the target signal characteristics, time step size information, and number of set center points; the initial population includes multiple initial individuals.

[0103] In this embodiment, the target signal features, the length of its time step, and the number of set center points can be extracted as gene values ​​for the first generation individuals, thereby generating multiple first generation individuals, which together form the first generation population.

[0104] Step S603: Calculate the signal-to-noise ratio based on the intermediate signal characteristics and the target signal characteristics.

[0105] In this embodiment, it is understood that target signal features are obtained by filtering intermediate signal features. These intermediate signal features are then used to remove noise and redundant information from the intermediate signal features, resulting in target signal features free of noise and redundant information. Therefore, the intermediate signal features contain noise and redundant information. The signal-to-noise ratio (SNR) can be calculated by dividing the target signal features by the difference between the target signal features and the intermediate signal features, and then taking the logarithm of the division result.

[0106] Step S604: Based on the target signal characteristics, the preset underwater target detection function, and the preset underwater target detection vector, the target to be detected is identified and analyzed, and the target recognition rate is calculated.

[0107] In this embodiment, the preset underwater target detection function and the preset underwater target detection vector can be manually set. The preset underwater target detection function maps target signal features into a high-dimensional nonlinear space, and then the preset underwater target detection vector performs a nonlinear transformation on the sound signal features in the high-dimensional nonlinear space. Finally, the nonlinearly transformed sound signal features are compared with preset nonlinear features of the target signal to calculate the target recognition rate.

[0108] Step S605: Calculate the fitness of each initial individual based on the signal-to-noise ratio, target recognition rate, and preset weight coefficients.

[0109] In this embodiment, the preset weighting coefficients can be set manually. They can be calculated by weighting and summing the signal-to-noise ratio and target recognition rate based on the preset weighting coefficients; the summation result represents the fitness of each initial individual.

[0110] Step S606: Based on the fitness of the first generation individuals, perform individual screening on the first generation population to generate the second generation population.

[0111] In this embodiment, the selection probability of each initial generation individual can be calculated based on its fitness. Then, the initial population is screened using these selection probabilities, and the screened individuals become the next generation individuals used to generate the next generation population. It is understood that individuals with higher fitness have a higher selection probability, and therefore a higher probability of becoming a next generation individual.

[0112] Step S607: Based on the preset crossover operator and the preset mutation operator, perform iterative calculations on the next generation population to obtain the last generation population.

[0113] In this embodiment, the preset crossover operator can be manually set and used to select pairs of individuals in the population with a certain probability, and randomly select certain gene values ​​for exchange. The preset mutation operator can also be manually set and used to transform the gene values ​​of a single individual with a certain probability. This can be based on Craig encoding to invert the binary bits in the gene values. The next generation population can be iterated through continuous crossover and mutation operations to optimize the individuals. The optimized individuals will then have their fitness recalculated, and individuals with high fitness will be selected as the last generation individuals to generate the final generation population.

[0114] Step S608: Determine the target time step size information and the number of target set center points based on the last generation population.

[0115] In this embodiment, the temporal step size and the number of set centroids corresponding to the individual with the highest fitness in the last generation population can be determined and output as target temporal step size information and target set centroid number information.

[0116] The signal feature extraction method provided in this application generates an initial population by using each target signal feature, its corresponding time step size, and the number of set centroids as initial individuals. Selection, crossover, and mutation calculations are performed on the individuals in the initial population to filter and optimize the time step size and the number of set centroids in various sound feature extraction schemes. By calculating the fitness of each individual, and then filtering the population individuals based on fitness, it avoids getting trapped in local optima during iteration and ensures that the iteration process is not disturbed by the initial population. This facilitates the rapid calculation of the global optimal solution, which is then output as the optimal value for the time step size and the number of set centroids in the sound feature extraction scheme. This allows for subsequent adjustments to the time step size and the number of set centroids, thereby improving the effectiveness and accuracy of sound feature extraction.

[0117] Figure 7 The flowchart illustrating the implementation of the signal feature extraction method provided in Embodiment 7 of this application is shown. The difference between this method and Embodiment 6 described above is that:

[0118] The preset underwater target detection function includes a preset feature space mapping function and preset Lagrange multipliers;

[0119] The preset underwater target detection vector includes a preset normal vector and a preset translation vector;

[0120] Step S604 specifically includes:

[0121] Step S701: The discrete feature values ​​of the target signal features are labeled to obtain multiple signal feature labels; the signal feature labels correspond one-to-one with the discrete feature values.

[0122] In this embodiment, it can be understood that the target signal features can be presented in the form of an array or sequence of multiple discrete values ​​with time-series correlation. All values ​​are extracted and their order in the array or sequence is marked to obtain multiple signal feature labels.

[0123] Step S702: Calculate the detection feature representation variables for each discrete feature value based on the signal feature label, discrete feature value, preset feature space mapping function, preset normal vector, preset translation vector, and preset Lagrange multiplier.

[0124] In this embodiment, the preset feature space mapping function, preset normal vector, and preset translation vector can be manually set, and the preset Lagrange multiplier can be designed based on the Lagrange function. The preset feature space mapping function can be a Gaussian radial basis function, used to map the target signal features to a high-dimensional feature space, making the nonlinear target signal feature values ​​linearly separable. The preset normal vector and preset translation vector are used to form a logic hyperplane to divide the target signal features into detection target feature values ​​and non-detection target feature values, characterizing whether the target signal features can accurately identify the target to be detected. The preset normal vector is used to characterize the direction of the logic hyperplane, and the preset translation vector is used to limit the boundary of the logic hyperplane. Specifically, each discrete feature value can be used as the independent variable of the feature space mapping function to calculate the feature space mapping function value. Then, the feature space mapping function is multiplied by the signal feature label, the normal vector, and the translation vector, and then processed by the Lagrange multiplier. The calculation result is the detection feature characterization variable, used for subsequent calculations to determine whether the sound signal features can accurately detect the target to be detected.

[0125] In this embodiment, preferably, the preset normal vector and the preset translation vector can be obtained through continuous optimization in the past research on target detection, or they can be obtained through continuous training of a specific calculation model. They are used to limit the generation range of the logical hyperplane, thereby achieving effective differentiation between target feature values ​​and non-detection target feature values.

[0126] Step S703: Determine whether the detected feature representation variable is greater than zero; if yes, proceed to step S704; if no, proceed to step S705.

[0127] In this embodiment, when the detected feature representation variable is greater than zero, it indicates that the sound signal feature can effectively represent the feature of the target to be detected, and thus the sound signal feature is a valid feature. When the detected feature representation variable is less than or equal to zero, it indicates that the sound signal feature cannot effectively represent the feature of the target to be detected, and thus the sound signal feature is an invalid feature. It is understood that even after filtering out noise or redundant information, the sound signal features still contain signal features that can effectively represent the information of the target to be detected, as well as signal features that cannot effectively represent the information of the target to be detected. Therefore, it is necessary to calculate the target recognition rate to quantify the effectiveness and usability of the extracted sound signal features, and to evaluate the applicability of the sound signal extraction scheme through the target recognition rate, thereby selecting sound signal extraction schemes that can extract more effective features.

[0128] Step S704: Determine the discrete feature values ​​corresponding to the detection feature characterization variables as the detection target feature values.

[0129] In this embodiment, when the detection feature representation variable is greater than zero, it indicates that the sound signal feature can effectively represent the feature of the target to be detected. Therefore, the discrete feature value corresponding to the detection feature representation variable is determined as the target feature value, which is used to calculate the target recognition rate in the subsequent process.

[0130] Step S705: Determine the discrete feature values ​​corresponding to the detection feature characterization variables as non-detection target feature values.

[0131] In this embodiment, when the detection feature representation variable is less than or equal to zero, it indicates that the sound signal feature cannot effectively represent the feature of the target to be detected. Therefore, the discrete feature value corresponding to the detection feature representation variable is determined as the non-target feature value, which is used for subsequent calculation of the target recognition rate.

[0132] Step S706: Calculate the target recognition rate based on the detected target feature values ​​and the non-detected target feature values.

[0133] In this embodiment, the target recognition rate can be obtained by dividing the number of detected target feature values ​​by the sum of the number of target feature values ​​and the number of non-detected target feature values, which is used to quantify the effectiveness of each sound signal feature.

[0134] The signal feature extraction method provided in this application determines the effectiveness of the extracted sound signals. Specifically, it calculates the optimal decision boundary using preset normal vectors, translation vectors, and Lagrange multipliers to maximize the classification margin. This allows for accurate differentiation between target feature values ​​and non-detection target feature values, thereby calculating the target recognition rate. This target recognition rate facilitates the detection of the effectiveness of the extracted sound signal features, enabling optimization and screening of sound signal extraction schemes. The optimal sound signal extraction scheme obtained through screening ensures the accuracy, effectiveness, and robustness of sound signal extraction in underwater target detection.

[0135] Corresponding to the method in the above embodiments, Figure 8 A structural block diagram of the signal feature extraction device provided in the embodiments of this application is shown. For ease of explanation, only the parts related to the embodiments of this application are shown. Figure 8 The example signal feature extraction device can be the execution subject of the signal feature extraction method provided in the aforementioned embodiment 1.

[0136] Reference Figure 8 The signal feature extraction device includes:

[0137] The underwater sound signal acquisition module 810 is used to acquire the underwater sound signal of the target to be detected.

[0138] The initial signal feature generation module 820 is used to extract time-series features from the underwater sound signal based on multiple preset time-series step-size information, preset state variables, preset gate weight matrix, preset gate bias variables and preset normalization function, and generate initial signal features.

[0139] The intermediate signal feature generation module 830 is used to enhance the initial signal features according to multiple preset linear transformation matrices to generate intermediate signal features; and

[0140] The target signal feature generation module 840 is used to classify and calculate the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function, and generate target signal features.

[0141] The process by which each module in the signal feature extraction device provided in this application implements its respective function can be found in the foregoing. Figure 1 The description of Embodiment 1 shown will not be repeated here.

[0142] The signal feature extraction device provided in this application embodiment implements a method that considers the autocorrelation and cross-correlation of each sound signal sequence. By performing preliminary feature extraction on the underwater sound information of the target to be detected, it simultaneously captures the time domain, frequency domain, and temporal features of the sound signal. Then, it performs linear temporal feature enhancement processing and classification processing on the preliminary extracted features, filters out noise and redundant information in the signal features, and improves the comprehensiveness, effectiveness, and accuracy of signal feature extraction. This enables comprehensive and effective extraction of sound signals in complex aquatic environments, thereby improving the accuracy and robustness of underwater target detection through the extracted sound signal features.

[0143] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0144] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0145] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0146] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0147] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. It should also be understood that although the terms "first," "second," etc., are used in the text to describe various elements in some embodiments of this application, these elements should not be limited by these terms. These terms are merely used to distinguish one element from another. For example, a first table may be named a second table, and similarly, a second table may be named a first table, without departing from the scope of the various described embodiments. Both the first table and the second table are tables, but they are not the same table.

[0148] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0149] The signal feature extraction method provided in this application can be applied to terminal devices such as mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, laptops, ultra-mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs). This application does not impose any restrictions on the specific type of terminal device.

[0150] For example, the terminal device may be a station (STAION, ST) in a WLAN, a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA) device, a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a vehicle networking terminal, a computer, a laptop computer, a handheld communication device, a handheld computing device, a satellite wireless device, a wireless modem card, a set-top box (STB), customer premises equipment (CPE), and / or other devices used for communication over a wireless system, as well as next-generation communication systems, such as mobile terminals in 5G networks or mobile terminals in future evolved Public Land Mobile Network (PLMN) networks.

[0151] As an example and not a limitation, when the terminal device is a wearable device, the term "wearable device" can also refer to any device that utilizes wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices worn directly on the body or integrated into a user's clothing or accessories. Wearable devices are not merely hardware devices; they achieve powerful functions through software support, data interaction, and cloud interaction. Broadly defined, wearable smart devices include those with comprehensive functions, large sizes, and the ability to perform complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those focused on a specific application function that require interaction with other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0152] Figure 9 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. For example... Figure 9 As shown, the terminal device 9 of this embodiment includes: at least one processor 90 ( Figure 9 (Only one is shown in the image) A memory 91 stores a computer program 92 that can run on the processor 90. When the processor 90 executes the computer program 92, it implements the steps in the various signal feature extraction method embodiments described above, for example... Figure 1 Steps S101 to S104 are shown. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 8The functions of modules 810 to 840 are shown.

[0153] The terminal device 9 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 90 and a memory 91. Those skilled in the art will understand that... Figure 9 This is merely an example of terminal device 9 and does not constitute a limitation on terminal device 9. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input transmission devices, network access devices, buses, etc.

[0154] The processor 90 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0155] In some embodiments, the memory 91 may be an internal storage unit of the terminal device 9, such as a hard disk or memory of the terminal device 9. The memory 91 may also be an external storage device of the terminal device 9, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device 9. Furthermore, the memory 91 may include both internal and external storage units of the terminal device 9. The memory 91 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 91 can also be used to temporarily store data that has been sent or will be sent.

[0156] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0157] This application also provides a terminal device, which includes at least one memory, at least one processor, and a computer program stored in the at least one memory and executable on the at least one processor. When the processor executes the computer program, it causes the terminal device to implement the steps in any of the above method embodiments.

[0158] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0159] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0160] If the integrated module / unit is implemented as 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, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0161] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0163] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0164] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A signal feature extraction method, characterized in that, include: Acquire underwater acoustic signals from the target to be detected; Based on multiple preset time step information, preset state variables, preset gate weight matrix, preset gate bias variable, and preset normalization function, the underwater sound signal is subjected to time feature extraction to generate initial signal features. The initial signal features are enhanced based on multiple preset linear transformation matrices to generate intermediate signal features. Based on multiple randomly generated set center points and a preset set spatial displacement function, the intermediate signal features are classified and calculated to generate target signal features; After the step of classifying and calculating the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function to generate target signal features, the method further includes: Extract the time step size information and the number of set center points corresponding to multiple target signal features; Based on the target signal characteristics, time step size information, and number of set center points, an initial population is generated; the initial population includes multiple initial individuals. Calculate the signal-to-noise ratio based on the intermediate signal characteristics and the target signal characteristics; Based on the target signal characteristics, the preset underwater target detection function, and the preset underwater target detection vector, the target to be detected is identified and analyzed, and the target recognition rate is calculated. The fitness of each initial individual is calculated based on the signal-to-noise ratio, target recognition rate, and preset weight coefficients. Based on the fitness of the first-generation individuals, the first-generation population is screened to generate the second-generation population; Based on preset crossover and mutation operators, the next generation population is iteratively calculated to obtain the last generation population. Based on the last generation population, determine the target time step size information and the number of target set center points; The preset underwater target detection function includes a preset feature space mapping function and preset Lagrange multipliers; The preset underwater target detection vector includes a preset normal vector and a preset translation vector; The steps of identifying and analyzing the target to be detected based on the target signal features, a preset underwater target detection function, and a preset underwater target detection vector, and calculating the target recognition rate, specifically include: The discrete feature values ​​of the target signal are labeled to obtain multiple signal feature labels; the signal feature labels correspond one-to-one with the discrete feature values. Based on the signal feature labels, discrete feature values, preset feature space mapping function, preset normal vector, preset translation vector, and preset Lagrange multipliers, calculate the detection feature representation variables for each discrete feature value. Determine whether the detected feature characterization variable is greater than zero; If so, the discrete feature values ​​corresponding to the detection feature characterization variables are determined as the detection target feature values; If not, the discrete feature values ​​corresponding to the detection feature characterization variables are determined as non-detection target feature values; The target recognition rate is calculated based on the detected target feature values ​​and the non-detected target feature values.

2. The signal feature extraction method as described in claim 1, characterized in that, The preset state variables include preset hidden state variables and preset updated state variables; The preset gating weight matrix includes a preset pre-gating weight matrix, a preset first intermediate gating weight matrix, a preset second intermediate gating weight matrix, and a preset post-gating weight matrix; The preset gating bias variables include preset pre-gating bias variables, preset first intermediate gating bias variables, preset second intermediate gating bias variables, and preset post-gating bias variables. The step of extracting temporal features from the underwater sound signal and generating initial signal features based on multiple preset time step sizes, preset state variables, preset gate weight matrices, preset gate bias variables, and preset normalization functions specifically includes: Based on multiple preset time step information, the underwater sound signal is segmented and spliced ​​to obtain multiple signal time sequence information. Based on the preset pre-gating weight matrix, preset hidden state variables, preset pre-gating bias variables, and preset standardization function, the time sequence information of multiple signal segments is mapped, transformed, and standardized to obtain the time sequence feature values ​​of the pre-gating signal. Based on the preset first intermediate gating weight matrix, the preset hidden state variables, the preset first intermediate gating bias variables, and the preset standardization function, the time sequence information of multiple signal segments is mapped, transformed, and standardized to obtain the time sequence feature value of the first intermediate gating signal. Based on the preset second intermediate gating weight matrix, the preset hidden state function, and the preset second intermediate gating bias variable, the signal timing segment information is mapped, transformed, and standardized to obtain intermediate gating signal timing feature variables. Perform a hyperbolic tangent transform on the timing characteristic variables of the intermediate gate signal to obtain the timing characteristic values ​​of the second intermediate gate signal; Based on the preset post-gating weight matrix, preset hidden state variables, preset post-gating bias variables, and preset standardization function, the time sequence information of multiple signal segments is mapped, transformed, and standardized to obtain the time sequence feature values ​​of the post-gating signal. The initial signal features are generated based on the timing feature values ​​of the pre-gating signal, the first intermediate gating signal, the second intermediate gating signal, and the post-gating signal.

3. The signal feature extraction method as described in claim 2, characterized in that, The step of generating initial signal features based on the timing feature values ​​of the pre-gated gate signal, the first intermediate gate signal, the second intermediate gate signal, and the post-gated gate signal specifically includes: Based on the timing characteristic values ​​of the pre-gated gate signal, the timing characteristic values ​​of the first intermediate gate signal, the timing characteristic values ​​of the second intermediate gate signal, and the preset update state variables, calculate the timing characteristic variables of the state representation. Perform a hyperbolic tangent transform on the state representation time-series characteristic variables to obtain the state transformation time-series characteristic variables; Based on the timing characteristic values ​​of the post-gated gate signal and the timing characteristic variables of the state transition, calculate multiple timing characteristic representation variables of the audio signal; Initial signal features are generated based on multiple time-series characteristic representation variables of the aforementioned sound signals.

4. The signal feature extraction method as described in claim 1, characterized in that, The preset linear transformation matrices include a preset first linear layer transformation matrix, a preset second linear layer transformation matrix, and a preset third linear layer transformation matrix; The step of enhancing the initial signal features based on multiple preset linear transformation matrices to generate intermediate signal features specifically includes: Based on the initial signal features, the preset first linear layer transformation matrix, the preset second linear layer transformation matrix, and the preset third linear layer transformation matrix, calculate the first linear layer feature representation variables, the second linear layer feature representation variables, and the third linear layer feature representation variables; Based on the feature representation variables of the first linear layer and the feature representation variables of the second linear layer, the intermediate linear representation feature variables are obtained; Multiply the intermediate linear representation feature variables and the third linear layer feature representation variables to obtain the signal feature linear representation variables; Based on the preset linear layer weight matrix, the linear representation variables of the signal features are weighted and summed to generate intermediate signal features.

5. The signal feature extraction method as described in claim 1, characterized in that, The step of classifying and calculating the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function to generate target signal features specifically includes: Calculate the distance between the intermediate signal features and the feature set representations of multiple set center points; Based on the feature set representation distance, the intermediate signal features are divided into multiple sets corresponding to multiple set center points to generate multiple signal feature sets; Based on the set of signal features and the preset set space displacement function, calculate the set space displacement of each intermediate signal feature; Based on the spatial displacement of the set, determine the displacement center point of each of the signal feature sets; Determine whether the displacement center point is the same as the set center point; If so, then target signal features are generated based on multiple sets of said signal features; If not, then the displacement center point is taken as the set center point, and the process returns to the step of dividing the intermediate signal features into sets corresponding to multiple set center points based on the feature set representation distance, thereby generating multiple signal feature sets.

6. A signal feature extraction device, characterized in that, include: The underwater sound signal acquisition module is used to acquire the underwater sound signal of the target to be detected. The initial signal feature generation module is used to extract time-series features from the underwater sound signal based on multiple preset time-series step-size information, preset state variables, preset gate weight matrix, preset gate bias variables, and preset normalization function, and generate initial signal features. The intermediate signal feature generation module is used to enhance the initial signal features based on multiple preset linear transformation matrices to generate intermediate signal features. as well as The target signal feature generation module is used to classify and calculate the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function, and generate target signal features. After the step of classifying and calculating the intermediate signal features based on multiple randomly generated set center points and a preset set spatial displacement function to generate target signal features, the method further includes: Extract the time step size information and the number of set center points corresponding to multiple target signal features; Based on the target signal characteristics, time step size information, and number of set center points, an initial population is generated; the initial population includes multiple initial individuals. Calculate the signal-to-noise ratio based on the intermediate signal characteristics and the target signal characteristics; Based on the target signal characteristics, the preset underwater target detection function, and the preset underwater target detection vector, the target to be detected is identified and analyzed, and the target recognition rate is calculated. The fitness of each initial individual is calculated based on the signal-to-noise ratio, target recognition rate, and preset weight coefficients. Based on the fitness of the first-generation individuals, the first-generation population is screened to generate the second-generation population; Based on preset crossover and mutation operators, the next generation population is iteratively calculated to obtain the last generation population. Based on the last generation population, determine the target time step size information and the number of target set center points; The preset underwater target detection function includes a preset feature space mapping function and preset Lagrange multipliers; The preset underwater target detection vector includes a preset normal vector and a preset translation vector; The steps of identifying and analyzing the target to be detected based on the target signal features, a preset underwater target detection function, and a preset underwater target detection vector, and calculating the target recognition rate, specifically include: The discrete feature values ​​of the target signal are labeled to obtain multiple signal feature labels; the signal feature labels correspond one-to-one with the discrete feature values. Based on the signal feature labels, discrete feature values, preset feature space mapping function, preset normal vector, preset translation vector, and preset Lagrange multipliers, calculate the detection feature representation variables for each discrete feature value. Determine whether the detected feature characterization variable is greater than zero; If so, the discrete feature values ​​corresponding to the detection feature characterization variables are determined as the detection target feature values; If not, the discrete feature values ​​corresponding to the detection feature characterization variables are determined as non-detection target feature values; The target recognition rate is calculated based on the detected target feature values ​​and the non-detected target feature values.

7. A terminal device, characterized in that, The terminal device includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images