Fiber optic hydrophone target event classification method, device, equipment and medium

Abstract

The present invention proposes a method, device, equipment and medium for classifying target events of optical fiber hydrophones, including inputting interference sampling signals obtained by sensing different known types of hydroacoustic target events with an optical fiber hydrophone, constructing optical fiber hydrophone sampling signal data sets corresponding to different types of hydroacoustic target events; inputting the optical fiber hydrophone sampling signal data sets into a neural network model for training using forward propagation and gradient descent methods to obtain a trained hydroacoustic target event classification detection model; based on the trained hydroacoustic target event classification detection model, performing hydroacoustic target event classification detection on the interference sampling signals obtained by sensing the hydroacoustic target events with the optical fiber hydrophone, and obtaining the classification results of the corresponding hydroacoustic target events. The present invention uses neural networks and machine learning to automatically extract features, realizes direct mapping from time-domain optical signals to target event types, simplifies the data processing flow, and optimizes the efficiency of target event detection.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of target event classification, in particular to a method, device, equipment and medium for classifying target events of an optical fiber hydrophone. Background Art

[0002] Sound is the only information carrier that can be transmitted over long distances in the marine environment. The analysis of target radiation noise characteristics based on underwater acoustic signal processing is the most important processing method in the field of underwater information reconnaissance and target detection. Fiber optic hydrophone is a type of underwater acoustic detection equipment based on fiber optic sensing. It has the advantages of high sensitivity and strong anti-interference ability. It is widely used in the field of underwater acoustic target detection and event detection.

[0003] Unlike piezoelectric hydrophones, the optoelectronic signal acquisition circuit of fiber optic hydrophones samples the optical signal in the sensor, while the sound signal is modulated in the phase and amplitude of the optical signal, and needs to be demodulated to recover the sound signal. Therefore, signal demodulation processing is an indispensable component of the hydroacoustic processing system based on fiber optic hydrophones, and the related demodulation processing circuit inevitably increases the complexity of the processing system and limits the miniaturization and integration of the system.

[0004] In recent years, artificial intelligence technology has developed rapidly, and intelligent processing algorithms such as machine learning and neural networks have been widely used in the field of underwater acoustic target event detection. Existing target classification methods are mainly based on machine learning and neural network training based on artificially designed and extracted features, such as spectral features of sound signals, short-time Fourier transform spectra, Mel energy spectra, and other spectral features. Since the above features are all obtained based on sound signal processing, this architecture cannot get rid of its dependence on the traditional demodulation circuit of the fiber optic hydrophone. The output optical signal of the fiber optic hydrophone needs to be demodulated and then feature extracted to obtain features that can be used for neural network training and reasoning.

[0005] In addition, when implementing and deploying existing intelligent target classification algorithms in real time, the implementation architectures of different computing steps such as signal demodulation, feature extraction, and neural network reasoning of fiber optic hydrophones vary greatly, and often require different types of hardware platforms for deployment and implementation. For example, real-time demodulation of fiber optic hydrophones generally requires FPGA to implement, while feature extraction such as time-frequency analysis requires CPU or DSP, and neural network reasoning requires GPU or AI acceleration card to implement. In summary, the existing intelligent systems for fiber optic hydrophone target event classification have problems such as complex implementation architecture and difficulty in algorithm deployment, which limits the application of artificial intelligence technology in the field of fiber optic hydrophones. Summary of the invention

[0006] In view of the problems existing in the prior art, the present invention provides a method, device, equipment and medium for classifying target events of a fiber optic hydrophone.

[0007] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0008] In one aspect, the present invention provides a method for classifying target events using a fiber optic hydrophone, comprising:

[0009] The interference sampling signals obtained by the fiber optic hydrophone sensing different known types of hydroacoustic target events are input to construct a fiber optic hydrophone sampling signal data set corresponding to different types of hydroacoustic target events;

[0010] The fiber optic hydrophone sampling signal data set is input into the neural network model and trained using the forward propagation and gradient descent method to obtain a trained underwater acoustic target event classification detection model;

[0011] Based on the trained underwater acoustic target event classification detection model, the interference sampling signal obtained by the optical fiber hydrophone sensing the underwater acoustic target event is subjected to underwater acoustic target event classification detection to obtain the corresponding underwater acoustic target event classification result.

[0012] In another aspect, a fiber optic hydrophone target event classification device is provided, comprising a 3×3 fiber optic hydrophone probe, a 3-channel optoelectronic acquisition module, a single-chip ZYNQ series FPGA, and a DDR cache;

[0013] The 3×3 fiber optic hydrophone probe uses an interferometric fiber optic hydrophone probe based on a 3×3 coupler to pick up the hydroacoustic signal and output 3 return interference optical signals;

[0014] The 3-channel photoelectric acquisition module is responsible for converting the 3-way returned interference light signals into electrical signals, and realizes analog-to-digital conversion through ADC, and outputs 3-way sampling data;

[0015] The single-chip ZYNQ series FPGA includes a 3-channel ADC interface module, a cache FIFO module, an APU module, an AXI interface module, and a DPU acceleration module; the 3-channel ADC interface module is responsible for receiving the digital signal output by the 3-channel photoelectric acquisition module outside the FPGA and converting the digital signal into data that can be processed inside the FPGA;

[0016] The cache FIFO module is responsible for caching the data output by the 3-channel ADC interface module to realize data clock domain conversion; the APU module is implemented by the ARM processor integrated inside the ZYNQ series FPGA, loaded with a trained underwater acoustic target event classification detection model, and is responsible for the process management, parameter loading and part of lightweight data calculation of the entire calculation process of classification detection based on the trained underwater acoustic target event classification detection model; the DPU acceleration module is implemented by the programmable logic resources inside the ZYNQ series FPGA, and uses programmable logic resources to implement the acceleration engine of neural network calculation, realizing hardware acceleration in the calculation process of classification detection based on the trained underwater acoustic target event classification detection model; the AXI data interaction module includes an AXI interface unit and a multi-channel DMA control unit, which is responsible for completing high-speed data interaction between the cache FIFO module, APU module, DPU acceleration module and external DDR cache;

[0017] The DDR cache is a large-capacity cache chip outside the FPGA chip, which is responsible for the large-capacity and high-speed data cache function in the entire calculation process of data processing and classification detection based on the trained underwater acoustic target event classification detection model.

[0018] On the other hand, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned fiber optic hydrophone target event classification method when executing the computer program.

[0019] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned fiber optic hydrophone target event classification method are implemented.

[0020] In another aspect, the present invention provides a computer program product, which is stored on a computer-readable storage medium and includes computer instructions, which, when executed by a processor, enable a computer device to implement the steps of the above-mentioned fiber optic hydrophone target event classification method.

[0021] Compared with the prior art, the technical effects of the present invention are:

[0022] The classification method provided by the present invention omits the optoelectronic signal demodulation process of the fiber optic hydrophone, directly starts from the optical signal domain, utilizes neural networks and machine learning to automatically extract features, realizes direct mapping from time domain optical signals to target events, simplifies the data processing process, and optimizes the efficiency of target event detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0024] Figure 1 A flowchart of a method for classifying target events using a fiber optic hydrophone provided by an embodiment;

[0025] Figure 2 is a structural block diagram of a neural network model used in an embodiment;

[0026] Figure 3 This is a model deployment diagram of an embodiment using a single-chip ZYNQ series FPGA to implement fiber optic hydrophone signal acquisition and rapid target detection. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Reference Figure 1 In one embodiment, a method for classifying target events of a fiber optic hydrophone is provided, comprising the following steps:

[0029] The interference sampling signals obtained by the fiber optic hydrophone sensing different known types of hydroacoustic target events are input to construct a fiber optic hydrophone sampling signal data set corresponding to different types of hydroacoustic target events;

[0030] The fiber optic hydrophone sampling signal data set is input into the neural network model and trained using the forward propagation and gradient descent method to obtain a trained underwater acoustic target event classification detection model;

[0031] Based on the trained underwater acoustic target event classification detection model, the interference sampling signal obtained by the optical fiber hydrophone sensing the underwater acoustic target event is subjected to underwater acoustic target event classification detection to obtain the corresponding underwater acoustic target event classification result.

[0032] The above classification method omits the optoelectronic signal demodulation process of the fiber optic hydrophone, starts directly from the optical signal domain, uses neural networks and machine learning to automatically extract features, realizes direct mapping from time domain optical signals to target event types, simplifies the data processing process, and optimizes the efficiency of target event detection.

[0033] Interference sampling signals obtained by sensing different known types of hydroacoustic target events with a fiber optic hydrophone, wherein the fiber optic hydrophone can be implemented using a traditional fiber optic hydrophone and a signal acquisition system. Preferably, the fiber optic hydrophone is an interferometric fiber optic hydrophone. For example, in a preferred embodiment, the interferometric fiber optic hydrophone is an interferometric fiber optic hydrophone based on a 3×3 coupler. An interferometric fiber optic hydrophone based on a 3×3 coupler is used to sense a variety of known types of hydroacoustic target events. The interference optical signal returned by the interferometric fiber optic hydrophone based on the 3×3 coupler is subjected to photoelectric detection and ADC sampling to obtain a fiber optic hydrophone sampling signal data set corresponding to different types of hydroacoustic events.

[0034] Specifically, the interferometric fiber optic hydrophone based on the 3×3 coupler senses the underwater acoustic target event. The interferometric optical signal returned by the interferometric fiber optic hydrophone is the interferometric optical signal of the three output arms of the 3×3 coupler. The three interferometric optical signals are collected by three photodetectors and ADC circuits respectively to obtain the sampling data of the three interferometric optical signals. , and , where three sampling data , and The expression is as follows:

[0035] ;

[0036] In the above formula, , and is the DC light intensity of the three output arms of the 3×3 coupler; , and is the interference phase correlation term of the three output arms of the 3×3 coupler; , and is the phase noise in the three output arms of the 3×3 coupler; is the optical path phase difference between the second output arm and the first output arm; is the optical path phase difference between the third output arm and the first output arm, It is the original signal corresponding to the hydroacoustic target event;

[0037] The length of the three paths is Sampling data of , and Constitute a The matrix constitutes the interference sampling signal obtained by the fiber optic hydrophone sensing the hydroacoustic target event, which serves as the input of the subsequent neural network.

[0038] In the process of inputting the fiber optic hydrophone sampling signal data set into the neural network model and using the forward propagation and gradient descent method for training, the loss function is set in any way, including but not limited to the loss function being a mean square error loss function or a cross entropy loss function. In a preferred embodiment, considering the non-uniformity of the underwater acoustic target time samples, the loss function in the process of inputting the fiber optic hydrophone sampling signal data set into the neural network model and using the forward propagation and gradient descent method for training is expressed as:

[0039] ;

[0040] in, is the standard cross entropy loss function, is the rebalancing compensation factor for unbalanced multi-classification prediction, As a hyperparameter, in the implementation example of the present invention, it is generally set to 0.1. The loss function after compensation by the rebalancing compensation factor is compatible with the characteristics of the cross entropy function and alleviates the problems of training imbalance and classification effect degradation caused by uneven number of sample classifications.

[0041] For the architecture of the neural network model, refer to Figure 2 In a preferred embodiment of the present invention, a neural network model is proposed, wherein the neural network model includes three input channels, three feature extraction convolutional neural networks, a Conv1D convolutional neural network, n residual convolutional neural networks, a concatenation layer, a self-attention connection layer and a fully connected layer;

[0042] Three input channels, namely a first input channel, a second input channel and a third input channel, the three input channels are used to input sampling data of three-path interference optical signals;

[0043] Three feature extraction convolutional neural networks, namely the first feature extraction convolutional neural network, the second feature extraction convolutional neural network, and the third feature extraction convolutional neural network, perform convolution calculations on the input data of the three input channels respectively, and realize filtering and one-dimensional feature extraction of the three input data respectively. The convolution kernel sizes of the three feature extraction convolutional neural networks are all greater than 1. The three feature extraction convolutional neural networks have the same structure, but do not share parameters.

[0044] The Conv1D convolutional neural network uses a convolution kernel of scale 1×1 for convolution operations to achieve cross-channel fusion of one-dimensional features output by three feature extraction convolutional neural networks, thereby reducing the feature dimension and the complexity of subsequent processing.

[0045] n residual convolutional neural networks are respectively the first residual convolutional neural network, the second residual convolutional neural network, ..., the nth residual convolutional neural network; each residual convolutional neural network uses convolution kernels of different scales and structures to perform convolution calculations, and extracts signal features at different time scales from the output signal of the Conv1D convolutional neural network, each residual convolutional neural network is configured with pooling and nonlinear activation calculations, and after calculations by the n residual convolutional neural networks, n different feature vectors are output; wherein the value of n is generally not less than the number of known categories of underwater acoustic target events in the fiber optic hydrophone sampling signal data set.

[0046] The concatenation layer concatenates and fuses the n different feature vectors output by the n residual convolutional neural networks to obtain a feature vector of higher dimension.

[0047] The self-attention connection layer uses the self-attention mechanism to calculate features, extract the long-distance dependencies between different time and space segments in the feature vector output by the splicing layer, and improve the feature correlation between multiple feature vectors and feature vectors in different time and space spans. The output result is a feature vector containing self-attention weights;

[0048] The fully connected layer uses a fully connected convolutional neural network to perform multi-layer convolution operations on the input feature vector containing the self-attention weights. The output classification structure of the fully connected layer is the category label of the underwater acoustic target event.

[0049] The neural network model provided in the above embodiment has the following innovations: it eliminates the calculations such as demodulation and manual feature extraction of the traditional method, and instead directly learns the original data of the fiber optic hydrophone through the above neural network model for training and reasoning. The advantage of this is that traditional demodulation and manual feature extraction require computing resources such as FPGA and DSP, which are different from computing resources such as NPU for machine learning. Directly using the above neural network model can simplify the computing resources when deploying the algorithm.

[0050] In order to realize the above-mentioned fiber optic hydrophone target event classification method, the present invention also proposes a fiber optic hydrophone target event classification device, which realizes the whole process of fiber optic hydrophone signal from data acquisition to hydroacoustic target event classification reasoning in a single ZYNQ series FPGA, greatly simplifies the implementation architecture of machine learning intelligent algorithms in the field of fiber optic hydrophones, and significantly improves the system integration.

[0051] Specifically, an optical fiber hydrophone target event classification device is provided in one embodiment, referring to Figure 3 , Figure 3A model deployment diagram for implementing fiber optic hydrophone signal acquisition and rapid target detection using a single-chip ZYNQ series FPGA, including a 3×3 fiber optic hydrophone probe, a 3-channel optoelectronic acquisition module, a single-chip ZYNQ series FPGA, and a DDR cache.

[0052] The 3×3 fiber optic hydrophone probe uses an interferometric fiber optic hydrophone probe based on a 3×3 coupler to pick up the hydroacoustic signal and output 3 return interference optical signals;

[0053] The 3-channel photoelectric acquisition module is responsible for converting the 3-way return interference light signal into an electrical signal, and realizes analog-to-digital conversion through ADC to output 3-way sampling data. Specifically, the 3-way return light signal, i.e., the interference light signal of the three output arms of the 3×3 coupler, is collected by three photoelectric detectors and ADC circuits to obtain the sampling data of the three-way interference light signal. , and , where three sampling data , and The expression is as follows:

[0054] ;

[0055] In the above formula, , and is the DC light intensity of the three output arms of the 3×3 coupler; , and is the interference phase correlation term of the three output arms of the 3×3 coupler; , and is the phase noise in the three output arms of the 3×3 coupler; is the optical path phase difference between the second output arm and the first output arm; is the optical path phase difference between the third output arm and the first output arm, It is the original signal corresponding to the hydroacoustic target event.

[0056] Based on the trained underwater acoustic target event classification detection model, the interference sampling signal obtained by the optical fiber hydrophone sensing the underwater acoustic target event is subjected to underwater acoustic target event classification detection to obtain the corresponding classification result of the underwater acoustic target event. The entire classification process is deployed and implemented in a single-chip ZYNQ series FPGA, which mainly includes a 3-channel ADC interface module, a cache FIFO module, an APU module, an AXI interface module, and a DPU acceleration module.

[0057] The 3-channel ADC interface module is responsible for receiving the digital signals output by the 3-channel photoelectric acquisition module outside the FPGA and converting the digital signals into data that can be processed inside the FPGA;

[0058] The cache FIFO module is responsible for caching the data output by the 3-channel ADC interface module to realize data clock domain conversion;

[0059] The APU module is implemented by the ARM processor integrated inside the ZYNQ series FPGA and is loaded with a trained underwater acoustic target event classification and detection model. It is mainly responsible for the process management, parameter loading and some lightweight data calculation functions of the entire calculation process of classification and detection based on the trained underwater acoustic target event classification and detection model.

[0060] The DPU acceleration module is implemented by the programmable logic resources inside the ZYNQ series FPGA. It uses programmable logic resources such as logic gates and DSPSlice to implement the acceleration engine for neural network calculations. It is mainly responsible for completing high-performance calculations such as matrix multiplication and cyclic accumulation. It can achieve hardware acceleration in the calculation process of classification and detection based on the trained underwater acoustic target event classification detection model through multi-channel parallel computing.

[0061] The AXI data interaction module mainly consists of an AXI interface unit and a multi-channel DMA control unit, which is responsible for completing high-speed data interaction between the cache FIFO module, APU module, DPU acceleration module and external DDR cache.

[0062] The DDR cache is a large-capacity cache chip outside the FPGA chip, which is responsible for the large-capacity and high-speed data cache function in the entire calculation process of data processing and classification detection based on the trained underwater acoustic target event classification detection model.

[0063] In the specific implementation, the compilation tool provided by the ZYNQ series FPGA manufacturer is used to compile the trained underwater acoustic target event classification detection model into an algorithm executable file that supports running on the ZYNQ FPGA architecture, and then the executable file is burned into the APU module. After the APU module is started, the algorithm executable file is loaded, and during the execution process, the high-performance computing tasks such as matrix multiplication and loop accumulation are assigned to the DPU acceleration module for execution through the AXI data interaction module. The calculation results of the DPU acceleration module are then returned to the APU module through the AXI data interaction module, and finally the process of classifying underwater acoustic target events is completed based on the interference optical signals returned by the three-way optical fiber hydrophone.

[0064] In another embodiment, a fiber optic hydrophone target event classification device is provided, comprising:

[0065] The first module is used to input interference sampling signals obtained by the fiber optic hydrophone sensing different types of hydroacoustic target events, and construct a fiber optic hydrophone sampling signal data set corresponding to different types of hydroacoustic target events;

[0066] The second module is used to input the fiber optic hydrophone sampling signal data set into the neural network model and train it using the forward propagation and gradient descent method to obtain a trained underwater acoustic target event classification detection model;

[0067] The third module is used to perform underwater acoustic target event classification detection on the interference sampling signal obtained by the optical fiber hydrophone sensing the underwater acoustic target event based on the trained underwater acoustic target event classification detection model, and obtain the corresponding underwater acoustic target event classification result.

[0068] The implementation methods of the above modules and the construction of the model can adopt the methods described in any of the above embodiments, which will not be repeated here.

[0069] On the other hand, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the fiber optic hydrophone target event classification method provided in any of the above embodiments are implemented. The computer device may be a server. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store sample data. The network interface of the computer device is used to communicate with an external terminal via a network connection.

[0070] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the fiber optic hydrophone target event classification method provided in any of the above embodiments are implemented.

[0071] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0072] Matters not covered by the present invention are known technologies.

[0073] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0074] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for classifying target events using a fiber optic hydrophone, characterized in that: include: The interference sampling signals obtained by sensing different known types of hydroacoustic target events with the fiber optic hydrophone are input, and a fiber optic hydrophone sampling signal data set corresponding to different types of hydroacoustic target events is constructed. The fiber optic hydrophone is an interference type fiber optic hydrophone based on a 3×3 coupler. The interference type fiber optic hydrophone senses the hydroacoustic target event. The interference optical signal returned by the interference type fiber optic hydrophone is the interference optical signal of the three output arms of the 3×3 coupler. The three interference optical signals are collected by three photoelectric detectors and ADC circuits respectively to obtain sampling data of the three interference optical signals. , and , where three sampling data , and The expression is as follows: In the above formula, , and is the DC light intensity of the three output arms of the 3×3 coupler; , and is the interference phase correlation term of the three output arms of the 3×3 coupler; , and is the phase noise in the three output arms of the 3×3 coupler; is the optical path phase difference between the second output arm and the first output arm; is the optical path phase difference between the third output arm and the first output arm, It is the original signal corresponding to the hydroacoustic target event; The length of the three paths is Sampling data of , and Constitute a The matrix of constitutes the interference sampling signal obtained by the optical fiber hydrophone sensing the hydroacoustic target event, which serves as the input of the subsequent neural network; The fiber optic hydrophone sampling signal data set is input into the neural network model and trained using the forward propagation and gradient descent method to obtain a trained underwater acoustic target event classification detection model; Based on the trained underwater acoustic target event classification detection model, the interference sampling signal obtained by the optical fiber hydrophone sensing the underwater acoustic target event is subjected to underwater acoustic target event classification detection to obtain the corresponding underwater acoustic target event classification result.

2. The optical fiber hydrophone target event classification method according to claim 1, characterized in that: The fiber optic hydrophone sampling signal data set is input into the neural network model and trained using the forward propagation and gradient descent method. The loss function constructed is: in, is the cross entropy loss function, is the rebalancing compensation factor for unbalanced multi-classification prediction, as a preset hyperparameter.

3. The optical fiber hydrophone target event classification method according to claim 2, characterized in that: Set to 0.

1.

4. The method for classifying target events using optical fiber hydrophone according to claim 2, characterized in that: The neural network model includes three input channels, three feature extraction convolutional neural networks, a Conv1D convolutional neural network, n residual convolutional neural networks, a splicing layer, a self-attention connection layer and a fully connected layer; The three input channels are used to input sampling data of three-way interference optical signals; The three feature extraction convolutional neural networks perform convolution calculations on the input data of the three input channels respectively, and implement filtering and one-dimensional feature extraction on the three input data respectively; The Conv1D convolutional neural network uses a convolution kernel of scale 1×1 to perform convolution operations to achieve cross-channel fusion of the one-dimensional features output by the three feature extraction convolutional neural networks; n residual convolutional neural networks use convolution kernels of different scales and structures to perform convolution calculations, respectively extracting signal features at different time scales from the output signals of the Conv1D convolutional neural network. Each residual convolutional neural network is configured with pooling and nonlinear activation calculations. After calculations by n residual convolutional neural networks, n different feature vectors are output; The concatenation layer concatenates and fuses the n different feature vectors output by the n residual convolutional neural networks to obtain a high-dimensional feature vector; The self-attention connection layer uses the self-attention mechanism to calculate features, extract the long-distance dependencies between different time and space segments in the feature vector output by the splicing layer, and improve the feature correlation between multiple feature vectors and feature vectors in different time and space spans. The output result is a feature vector containing self-attention weights; The fully connected layer uses a fully connected convolutional neural network to perform multi-layer convolution operations on the input feature vector containing self-attention weights, and the output result is the category label of the underwater acoustic target event.

5. A fiber optic hydrophone target event classification device, characterized in that: Includes 3×3 fiber optic hydrophone probe, 3-channel optoelectronic acquisition module, single-chip ZYNQ series FPGA and DDR cache; The 3×3 fiber optic hydrophone probe uses an interferometric fiber optic hydrophone probe based on a 3×3 coupler to pick up the hydroacoustic signal and output 3 return interference optical signals; The 3-channel photoelectric acquisition module is responsible for converting the 3-way returned interference light signals into electrical signals, and realizes analog-to-digital conversion through ADC, and outputs 3-way sampling data; The single-chip ZYNQ series FPGA includes a 3-channel ADC interface module, a cache FIFO module, an APU module, an AXI interface module, and a DPU acceleration module; the 3-channel ADC interface module is responsible for receiving the digital signal output by the 3-channel photoelectric acquisition module outside the FPGA and converting the digital signal into data that can be processed inside the FPGA; The cache FIFO module is responsible for caching the data output by the 3-channel ADC interface module to realize data clock domain conversion; the APU module is implemented by the ARM processor integrated inside the ZYNQ series FPGA, loaded with the trained underwater acoustic target event classification detection model as described in claim 1, and is responsible for the process management, parameter loading and part of lightweight data calculation of the entire calculation process of classification detection based on the trained underwater acoustic target event classification detection model; the DPU acceleration module is implemented by the programmable logic resources inside the ZYNQ series FPGA, and uses programmable logic resources to implement the acceleration engine of neural network calculation, so as to realize hardware acceleration in the calculation process of classification detection based on the trained underwater acoustic target event classification detection model; the AXI data interaction module includes an AXI interface unit and a multi-channel DMA control unit, which is responsible for completing the high-speed data interaction between the cache FIFO module, the APU module, the DPU acceleration module and the external DDR cache; The DDR cache is a large-capacity cache chip outside the FPGA chip, which is responsible for the large-capacity and high-speed data cache function in the entire calculation process of data processing and classification detection based on the trained underwater acoustic target event classification detection model.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the fiber optic hydrophone target event classification method according to claim 1 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fiber optic hydrophone target event classification method according to claim 1 are implemented.

Images