Unmanned underwater vehicle stealth assessment system and method

Abstract

The present invention relates to an unmanned underwater vehicle concealment assessment system and method, belonging to the field of data processing technology, comprising obtaining hydrological data through a sensor carried by the unmanned underwater vehicle, obtaining time domain data through sonar detection of hydroacoustic signals, and obtaining two-dimensional time-frequency data after calculating the time domain data through a time-frequency analysis algorithm; inputting the two-dimensional time-frequency data into a convolutional neural network model to identify the type of target equipment; when the type of the target equipment is identified as a dynamic non-cooperative target, by obtaining the position of the unmanned underwater vehicle, calculating the broadband sound propagation loss in parallel through a sound propagation model according to the hydrological data, and calculating the concealment of the unmanned underwater vehicle, and evaluating the concealment of the unmanned underwater vehicle through the concealment; being able to quickly identify dynamic non-cooperative targets in the navigation area and quickly calculate the broadband sound propagation loss, and finally obtaining the environmental concealment assessment level, so that the unmanned underwater vehicle has the ability to quickly and autonomously assess the concealment of the environment in which it is located.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a stealth assessment system and method for an unmanned underwater vehicle. Background Art

[0002] As a small and medium-sized marine equipment, unmanned submersibles have outstanding advantages in terms of mobility, versatility, endurance, and ease of deployment. They can perform various types of tasks through the various sensors they carry, especially special tasks in harsh and dangerous environments, and are more economical than manned ships. Unmanned submersibles are an important emerging underwater work platform, an important component of the maritime unmanned system platform, and an important direction for countries to develop underwater equipment.

[0003] The navigation safety of unmanned submersibles, that is, the concealment and safety during diving, is an important research direction. Unmanned submersibles are smaller in size than manned ships such as submarines and are not easy to be observed; manned ships are equipped with active and passive sonar, and sonar operators monitor the sonar (including timbre, beats, and fluctuations) to determine the target. Limited by the size of the fuselage, the energy reserves of existing unmanned submersible designs are generally insufficient to support the processor chips with powerful computing power. Existing technologies cannot efficiently conduct concealment assessments and can only rely on signals obtained by passive sonar for automatic judgment. Without manual assistance, lightweight and fast concealment assessment methods can only be developed based on the characteristics of unmanned submersibles. In layman's terms, manned ships are more advanced and have manual decision-making; while unmanned submersibles need to rely on automated algorithms based on lower-level equipment to achieve concealment assessments. Summary of the invention

[0004] In view of the shortcomings existing in the relevant technologies, the purpose of the present invention is to provide an unmanned underwater vehicle stealth assessment system and method to solve the technical problem that the energy reserve of the existing technology design is generally insufficient to support the installation of a processor chip with powerful computing power, and the stealth assessment cannot be performed efficiently. It can only rely on signals obtained by passive sonar for automatic judgment without human assistance, and needs to rely on automated algorithms based on lower-level equipment to achieve stealth assessment.

[0005] The present invention provides an unmanned underwater vehicle stealth assessment system, comprising:

[0006] Data acquisition module: obtains hydrological data through sensors carried by unmanned submersibles, obtains time domain data through sonar detection of hydroacoustic signals, and obtains two-dimensional time-frequency data after calculating the time domain data through time-frequency analysis algorithms;

[0007] Target recognition module: input the two-dimensional time-frequency data into a convolutional neural network model to identify the type of target device;

[0008] Propagation loss calculation module: when the type of the target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained by a positioning device carried by the unmanned underwater vehicle, and broadband sound propagation loss is calculated in parallel by a sound propagation model according to the hydrological data;

[0009] Concealment evaluation module: calculating the concealment of the unmanned underwater vehicle according to the type of the target device and the broadband sound propagation loss, and evaluating the concealment of the unmanned underwater vehicle according to the concealment;

[0010] The target recognition module further includes: the compression and expansion module of the convolutional neural network model is configured to include: a compression layer and an expansion layer, the compression layer uses a convolution kernel of a 1×1 convolution kernel to reduce the number of parameters and the number of input channels, the expansion layer uses a 1×1 convolution kernel and a 3×3 convolution kernel, which are spliced ​​after convolution calculation respectively, and the multi-size convolution kernel performs convolution calculation to retain more feature information;

[0011] A skip connection layer is provided between adjacent compression and expansion modules, for introducing the output of the network layer before the skip connection layer into the network layer after the skip connection layer;

[0012] The convolutional layer of the convolutional neural network model adopts depthwise separable convolution to replace the standard 3×3 convolution operation in the compression and expansion module.

[0013] The embodiment of the present invention can quickly identify dynamic non-cooperative targets in the navigation area and quickly calculate the broadband sound propagation loss through the convolutional neural network model, and finally obtain the concealment assessment level of the environment, so that it has the ability to quickly and autonomously assess the concealment of the environment, providing a basis and support for subsequent related embodied intelligent algorithms.

[0014] In some embodiments of the present invention, the propagation loss calculation module further includes:

[0015] When the unmanned underwater vehicle identifies a dynamic non-cooperative target, it automatically selects the Kraken sound propagation model or the Bellhop3D sound propagation model according to the position, and uses multi-process parallel calculation to calculate the propagation loss in the cylindrical ocean area with radius R and depth Z centered on its own position. The calculation process is an N×2D weak three-dimensional approximation method: the cylindrical area to be calculated starts from an azimuth of 0° and is divided into two parts according to the azimuth. n °Equally divided into 360 / n Two-dimensional plane, calculate the depth Z , the distance is R The propagation loss of each two-dimensional plane is combined and equivalent to the propagation loss of the entire cylindrical area.

[0016] When the sea depth at the location is less than or equal to a preset depth, the broadband sound propagation loss is calculated using the Kraken model;

[0017] Otherwise, the broadband sound propagation loss is calculated using the Bellhop3D model;

[0018] Among them, the Kraken model is parallelized at the frequency level through MPI parallel programming technology. Assume that the number of cores used for calculation by the processor is m , indicating that it can be run m Process, broadband sound propagation frequency range a - b Hz, step size c Hz, the broadband ( b - a ) / c +1 frequency calculation task is evenly distributed to the above processes:

[0019] when m ≥( b - a ) / c+ 1 o'clock, before b - a ) / c +1 process each gets a frequency calculation task;

[0020] when m <( b - a ) / c +1, before ((( b - a ) / c +1) mod m ) processes get (( b - a ) / c +1) / m +1 frequency calculation task, the remaining processes get (( b - a ) / c +1) / m A frequency calculation task.

[0021] Each process calculates the frequency according to the assigned frequency calculation task, modifies the corresponding parameters in the environment configuration file (*.env) according to the current sea depth and number of layers, calls the Kraken_mpi.exe executable program to calculate the wave number, and then calculates the complex sound pressure of the sound field at the current frequency. When all frequency calculation tasks of all processes are completed, each process calls the MPI_Gather function to gather all calculation results to process No. 0 to obtain the three-dimensional complex sound pressure, and then uses the Kraken model to calculate the broadband sound propagation loss. The formula is as follows:

[0022]

[0023]

[0024] in,TL B is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; r is the distance dimension; z is the depth dimension; is the broadband sound pressure in a two-dimensional plane;

[0025] The wave number is obtained by calculation, and the complex sound pressure in the two-dimensional plane is calculated by combining the sea depth stratification and distance stratification, which is equivalent to 360 / n The complex sound pressure in a two-dimensional plane.

[0026] Among them, the Bellhop3D model is parallelized at the frequency and orientation levels through MPI parallel programming technology. Assume that the number of cores available for calculation is m , indicating that it is executable m Process, broadband sound propagation frequency range a - b Hz, step size c Hz, the broadband ( b - a ) / c +1 frequency of 360 / n The position calculation tasks are evenly distributed to the above processes:

[0027] when m ≥(( b - a ) / c+ 1)×(360 / n ) time, before (( b - a ) / c+ 1)×(360 / n ) processes each get a computing task;

[0028] When m<(( b - a ) / c+ 1)×(360 / n ) time, before (((( b - a ) / c+ 1)×(360 / n )) mod m ) processes get ((( b - a ) / c+ 1)×(360 / n )) / m +1 computing task, the remaining processes get ((( b - a ) / c+ 1)×(360 / n )) / m A position calculation task;

[0029] The trajectory of the sound line in the Bellhop3D model is not only related to the pitch angle during propagation. α Related, and no longer fixed to azimuth β , affected by factors such as terrain, horizontal refraction may occur. Each process is opened up in a certain direction calculation task (360 / n ) / m A two-dimensional array space;

[0030] Each process calculates the assigned azimuth task according to the terrain and ocean environment of the current location, modifies the corresponding parameters in the environment configuration file (*.env), calls the Bellhop3D_mpi.exe executable program to calculate the sound line propagation trajectory, and then calculates the complex sound pressure of the sound field at the current azimuth. When all frequency calculation tasks of all processes are completed, each process calls the MPI_Gather function to gather all calculation results to process No. 0 to obtain the three-dimensional complex sound pressure, and then uses the Bellhop3D model to calculate the broadband sound propagation loss. The formula is as follows:

[0031]

[0032]

[0033] in, TL B is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; t is the direction dimension; r is the distance dimension; z is the depth dimension.

[0034] The embodiment of the present invention calculates the sound propagation environment at different depths and adopts a multi-process parallel computing method, which can effectively save computing time and quickly provide data such as broadband sound propagation loss in nearby waters for subsequent evaluation of the unmanned underwater vehicle's own stealth.

[0035] In some embodiments of the present invention, the concealment assessment module further comprises:

[0036] If the dynamic non-cooperative target is threatening, the detection figure of merit is calculated according to whether the target device is equipped with an active sonar, and the concealment degree is determined according to the relationship between the detection figure of merit and the broadband sound propagation loss;

[0037] If the dynamic non-cooperative target is not threatening, the concealment degree is 1;

[0038] When the target device is equipped with an active sonar, the detection figure of merit is calculated using the active sonar equation, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined.

[0039] When the detection figure of merit is less than twice the broadband sound propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0;

[0040] When the target device is not equipped with an active sonar, the passive sonar equation is used to calculate the detection figure of merit, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined.

[0041] When the detection figure of merit is less than the broadband sound propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0;

[0042] Among them, the calculation model of the active sonar equation is:

[0043]

[0044] in, The figure of merit for detection when equipped with active sonar; SL is the active sonar source level; TS is the target intensity; NL is the ocean noise level; DI For directionality; DT For testing valve;

[0045] Wherein, the calculation model of the passive sonar equation is:

[0046]

[0047] in, is the detection figure of merit when no active sonar is carried; SL is the noise source level of the unmanned underwater vehicle radiation; NL is the ocean noise level; DI For directionality; DT For the detection valve.

[0048] The embodiment of the present invention can calculate the detection merit of the target device according to whether it is equipped with active sonar, and obtain the current concealment of the unmanned underwater vehicle according to the active and passive sonar equations, thereby providing a basis and support for the implementation of subsequent task formulation, path planning, emergency avoidance and other algorithms.

[0049] Some embodiments of the present invention further provide a method for evaluating the stealth of an unmanned underwater vehicle, comprising the following steps:

[0050] Data acquisition step: acquiring hydrological data through sensors carried by the unmanned underwater vehicle, acquiring time domain data through sonar detection of hydroacoustic signals, and obtaining two-dimensional time-frequency data after calculating the time domain data through a time-frequency analysis algorithm;

[0051] Target identification step: inputting the two-dimensional time-frequency data into a convolutional neural network model to identify the type of target device;

[0052] Propagation loss calculation step: when the type of the target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained by a positioning device carried by the unmanned underwater vehicle, and broadband sound propagation loss is calculated in parallel by a sound propagation model according to the hydrological data;

[0053] Concealment evaluation step: calculating the concealment of the unmanned underwater vehicle according to the type of the target device and the broadband sound propagation loss, and evaluating the concealment of the unmanned underwater vehicle by the concealment;

[0054] The target recognition step further includes: the compression and expansion module of the convolutional neural network model is configured to include: a compression layer and an expansion layer, the compression layer uses a convolution kernel of a 1×1 convolution kernel to reduce the number of parameters and the number of input channels, the expansion layer uses a 1×1 convolution kernel and a 3×3 convolution kernel, which are spliced ​​after convolution calculation respectively, and the multi-size convolution kernel performs convolution calculation to retain more feature information;

[0055] A skip connection layer is provided between adjacent compression and expansion modules, for introducing the output of the network layer before the skip connection layer into the network layer after the skip connection layer;

[0056] The convolutional layer of the convolutional neural network model adopts depthwise separable convolution to replace the standard 3×3 convolution operation in the compression and expansion module.

[0057] In some embodiments of the present invention, the propagation loss calculation step further includes:

[0058] When the sea depth of the location is less than or equal to the preset depth, the broadband sound propagation loss is calculated by the Kraken model;

[0059] Otherwise, the broadband sound propagation loss is calculated using the Bellhop3D model;

[0060] Among them, the formula for calculating broadband sound propagation loss using the Kraken model is as follows:

[0061]

[0062]

[0063] in, is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; r is the distance dimension; z is the depth dimension; is the broadband sound pressure in a two-dimensional plane;

[0064] Among them, the formula for calculating broadband sound propagation loss through the Bellhop3D model is as follows:

[0065]

[0066]

[0067] in, is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; t is the direction dimension; r is the distance dimension; z is the depth dimension.

[0068] In some embodiments of the present invention, the concealment assessment step further comprises:

[0069] If the dynamic non-cooperative target is threatening, the detection figure of merit is calculated according to whether the target device is equipped with an active sonar, and the concealment degree is determined according to the relationship between the detection figure of merit and the broadband sound propagation loss;

[0070] If the dynamic non-cooperative target is not threatening, the concealment degree is 1;

[0071] When the target device is equipped with an active sonar, the detection figure of merit is calculated using the active sonar equation, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined.

[0072] When the detection figure of merit is less than twice the broadband sound propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0;

[0073] When the target device is not equipped with an active sonar, the passive sonar equation is used to calculate the detection figure of merit, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined.

[0074] When the detection figure of merit is less than the broadband sound propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0;

[0075] Among them, the calculation model of the active sonar equation is:

[0076]

[0077] in, The figure of merit for detection when equipped with active sonar; SL is the active sonar source level; TS is the target intensity; NL is the ocean noise level; DI For directionality; DT For testing valve;

[0078] Wherein, the calculation model of the passive sonar equation is:

[0079]

[0080] in, is the detection figure of merit when no active sonar is carried; SL is the noise source level of the unmanned underwater vehicle radiation; NL is the ocean noise level; DI For directionality; DT For the detection valve.

[0081] The embodiments of the present invention can quickly identify dynamic non-cooperative targets in the navigation area and quickly calculate the broadband sound propagation loss, and finally obtain the stealth assessment level of the environment, so that it has the ability to quickly and autonomously assess the stealth of the environment, providing a basis and support for subsequent related embodied intelligent algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0083] Figure 1 A structural diagram of an unmanned underwater vehicle stealth assessment system provided by an embodiment of the present invention;

[0084] Figure 2 A schematic diagram of a stealth assessment system for an unmanned underwater vehicle provided in an embodiment of the present invention;

[0085] Figure 3 A convolutional neural network model structure diagram provided for an embodiment of the present invention;

[0086] Figure 4 A structural diagram of a compression and expansion module in a convolutional neural network model provided in an embodiment of the present invention;

[0087] Figure 5 A schematic diagram of a skip connection structure in a convolutional neural network model provided in an embodiment of the present invention;

[0088] Figure 6 A schematic diagram of a depthwise separable convolution in a convolutional neural network model provided by an embodiment of the present invention;

[0089] Figure 7 A parallel calculation flow chart of a propagation loss calculation module provided in an embodiment of the present invention;

[0090] Figure 8 A schematic diagram of the division of a weak three-dimensional approximate calculation method provided by an embodiment of the present invention;

[0091] Figure 9 A flow chart of a method for evaluating the stealth of an unmanned underwater vehicle provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0093] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0094] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0095] The technical solution of the present invention is described in detail below in conjunction with specific embodiments and the accompanying drawings.

[0096] As attached Figure 1 - 2 As shown, the present invention provides a stealth assessment system for an unmanned underwater vehicle, comprising:

[0097] Data acquisition module: obtain time domain data by detecting the hydroacoustic signal carried by the sonar on the unmanned submersible, and obtain two-dimensional time-frequency data after calculating the time domain data through the time-frequency analysis algorithm; optionally, the hydrological data is temperature, salinity, depth, ocean noise level, etc.; optionally, the time-frequency analysis algorithm is short-time Fourier transform STFT , the formula is as follows:

[0098]

[0099] In the formula, s ( t ) is the original time domain signal, w ( t - τ ) is a window function; preferably, the window function is a Hamming window; τ The center position of the current window; ω is the frequency; j is an imaginary unit;

[0100] Target recognition module: inputting the two-dimensional time-frequency data into the convolutional neural network model to identify the type of the target device; preferably, the convolutional neural network model is an improved SqueezeNet network model;

[0101] Propagation loss calculation module: when the type of target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained through the positioning device carried by the unmanned underwater vehicle, and the broadband sound propagation loss is calculated in parallel through the sound propagation model according to the position; optionally, the dynamic non-cooperative target is an enemy submarine, a manned ship, a medium and large unmanned equipment, a passenger ship, and a cruise ship; among them, the opposite of the dynamic non-cooperative target is a static target and a cooperative target: the cooperative target includes one's own unmanned underwater vehicle, one's own submarine, and a manned ship; the static target includes a buoy.

[0102] Concealment evaluation module: Calculate the concealment of the UUV based on the type of target equipment and broadband acoustic propagation loss, and evaluate the concealment of the UUV through the concealment;

[0103] The target recognition module further includes: Figure 3 - 4 As shown, the compression expansion module of the convolutional neural network model is configured to include:

[0104] Compression layer and expansion layer, the compression layer uses a 1×1 convolution kernel to reduce the number of parameters and the number of input channels, where the number of parameters is only 1 / 9 of the 3×3 convolution kernel; the expansion layer uses a 1×1 convolution kernel and a 3×3 convolution kernel, which are spliced ​​by channel after convolution calculation respectively, and multi-size convolution kernels perform convolution calculations to retain more feature information and enhance the feature extraction capability of the model, thereby improving the accuracy of the model; optionally, the improved SqueezeNet model includes a total of 8 compression and expansion modules, represented as compression and expansion module 1 to compression and expansion module 8, where a convolution layer and a maximum pooling layer are connected before the compression and expansion module 1, a maximum pooling layer is connected after the compression and expansion module 3 and the compression and expansion module 7, and a convolution layer and an average pooling layer are connected after the compression and expansion module 8.

[0105] Combined with Figure 5 As shown, a jump connection layer is provided between adjacent compression and expansion modules, that is, a jump connection is introduced before and after the compression and expansion modules 2, 4, 6, and 8 to introduce the output of the network layer before the jump connection layer into the network layer after the jump connection layer. By additionally introducing the features of the previous network layer, the feature loss in the convolution process is reduced, the gradient disappearance problem when the number of network layers is deep is alleviated, and back propagation is facilitated to speed up the training process. The output formula of the convolution layer after the introduction of the jump connection is:

[0106]

[0107] in, y n ( x ) is the first n The output of the convolutional layer; x is the input of the model; y n-1 ( x )for n -The output of 1 convolutional layer; F n ( x ) is the first n The output of the convolutional layer.

[0108] In order to increase the training speed, a batch normalization layer is added after the convolutional layer to normalize the data according to the mean and variance.

[0109] Combined with reference Figure 6As shown in the figure, the convolution layer of the improved SqueezeNet convolutional neural network model adopts depthwise separable convolution to replace the standard 3×3 convolution operation in the compression and expansion module; in the standard convolution, several multi-channel convolution kernels are used to process the input multi-channel image, and the output feature map extracts channel features and spatial features. The depthwise separable convolution separates the standard convolution operation from the two dimensions of space and channel, and is divided into two steps: channel-by-channel convolution and point-by-point convolution:

[0110] In the channel-by-channel convolution process, each channel of the input has and only has one channel number of 1 and size K × K The convolution kernel is responsible for convolution, and finally obtains a feature map with the same number of input channels, but it cannot be expanded and the channel information is not effectively utilized;

[0111] In the point-by-point convolution process, a 1×1 convolution kernel is used to perform weighted combination on the feature map obtained in the previous step to form a new feature map, thereby changing the number of output channels and integrating channel dimension information. For example, if the number of input channels is C 1 , the number of output channels is C 2 , the convolution kernel size is K × K , the number of parameters of the standard convolution is K × K × C 1 × C 2 , if we use a deep convolutional neural network, we only need K × K × C 1 +1× C 1 × C 2 parameters; Depthwise separable convolution can significantly reduce the number of parameters and has higher computational efficiency; Optionally, the training model uses the Pytorch framework, and the application model uses the PyTorch C++ interface - libtorch framework;

[0112] During the model training process, the open source Deepship dataset and the sea trial measured dataset were used for training: In the preprocessing stage, Matlab was used to crop the acoustic signal of the dataset to a length of 1024 points, and the blank segments were removed. The training labels were set to 32 categories ranging from 0 to 31 according to the target type. Then, the short-time Fourier transform was used to process the acoustic signal, in which the Hamming window length was set to 128 and the number of fft points was set to 256. The final two-dimensional time-frequency data size was 129×129×1 (width×height×number of channels). The data size after processing by each network layer is shown in Table 1, and the accuracy of the verification set exceeded 90%.

[0113] Table 1 Data processing of each layer of convolutional neural network

[0114]

[0115] The improved SqueezeNet convolutional neural network model can quickly identify dynamic non-cooperative targets in the navigation area and quickly calculate the broadband sound propagation loss, and finally obtain the stealth assessment level of the environment, enabling it to quickly and autonomously assess the stealth of the environment, providing the basis and support for subsequent related embodied intelligent algorithms.

[0116] Combined with reference Figure 7 As shown, in order to enable the unmanned underwater vehicle to automatically select the parallel Kraken sound propagation model for broadband propagation loss calculation or the parallel Bellhop3D sound propagation model for incoherent propagation loss calculation according to the location and depth information when identifying the dynamic non-cooperative target; optionally, the depth is 200 meters; optionally, the Kraken sound propagation model and the Bellhop3D sound propagation model are both in Fortran language, compiled into executable files and then called by mixed programming in C++;

[0117] Use multi-process parallel calculation of broadband sound propagation loss in a cylindrical ocean area with its own location as the center, radius, and depth, and simplify the calculation process to an N×2D weak three-dimensional approximation method: start the cylindrical area to be calculated from an azimuth of 0° and calculate it according to the azimuth. n °Equally divided into 360 / n A two-dimensional plane, where n Usually 1 or 0.1 is taken, and the propagation loss of each two-dimensional plane is calculated and combined, which is equivalent to the propagation loss of the entire cylindrical area. The temperature, salinity, depth data and longitude and latitude position information of the ocean environment are obtained through the sensors and positioning devices carried by the unmanned submersible. According to the 14-term empirical formula applicable to the ocean area except the seabed solitary hot salt well as shown in the following formula, the sound speed at the location is calculated and stored:

[0118]

[0119] in, SSP is the speed of sound (m / s); T is the seawater temperature (℃); S is the seawater salinity (‰); Z is the depth (m); Φ is the latitude (°); R is the radius; Z For depth.

[0120] According to the different water depths, the multi-process parallel computing module of the mixed sound field propagation model is divided into the following two cases:

[0121] When the sea depth at the acquired location is less than or equal to the preset depth, the broadband sound propagation loss is calculated using the Kraken model;

[0122] Among them, the change of the amplitude and phase of the acoustic signal during the propagation of the ocean sound field is studied through the simple normal wave theory. Under the condition of layered medium, the Kraken model uses finite difference to decompose the simple normal wave equation, which can obtain a fast and accurate solution. Considering the steady-state sound field generated by a simple harmonic sound source, assuming that the ocean channel is a cylindrically symmetric layered medium, the simple normal wave solution is an integral solution of the wave equation. Each simple normal wave independently satisfies the wave equation and boundary conditions and propagates at its own speed. The solution of the wave equation can be written as the distance function φ ( r ) and depth function ψ ( z ), that is,

[0123]

[0124] in, is the sound pressure, is the horizontal distance, For depth.

[0125] The Kraken model will have a depth of Z The seawater is divided into equally spaced M Layers, each layer spacing is Z / M , the continuous problem in the simple normal wave equation can be simplified to a standard eigenvalue problem using the finite difference approximation method, and the solution of the wave equation can be obtained:

[0126]

[0127] in, ρ is the density of seawater, z s is the sound source depth, Ψ ( z s , rl ) is a constant.

[0128] The Kraken model simplifies the ocean into a hard seabed. The sound field environment is independent of distance and terrain. The wave number is obtained by calculation. Combined with the sea depth stratification and distance stratification, the complex sound pressure on the two-dimensional plane is calculated, which is equivalent to 360 / n The complex sound pressure in a two-dimensional plane;

[0129] The Kraken model is parallelized at the frequency level using MPI parallel programming technology. Assume that the number of cores used for calculation is m , indicating that it can be run m Process, broadband sound propagation frequency range a - b Hz, step size c Hz, the broadband (b - a ) / c +1 frequency calculation task is evenly distributed to the above processes:

[0130] when m ≥( b - a ) / c+ 1 o'clock, before (b - a ) / c +1 process each gets a frequency calculation task;

[0131] when m <( b - a ) / c +1, before ((( b - a ) / c +1) mod m ) processes get (( b - a ) / c +1) / m +1 frequency calculation task, the remaining processes get (( b - a ) / c +1) / m frequency calculation tasks. For example, if the number of cores available for calculation is 8, the number of processes that can be run is 8, and the frequency range of broadband sound propagation is 10-15 Hz:

[0132] When the step size is 1Hz, there are 6 frequency calculation tasks (15-10) / 1+1, which is less than the total number of processes (8). Then the first 6 processes 0-5 each get 1 frequency calculation task, and processes 6-7 are idle.

[0133] When the step size is 0.1Hz, there are (15-10) / 0.1+1, a total of 51 frequency calculation tasks, which is more than the total number of processes 8. Then the first three processes 0-2 each get 7 frequency calculation tasks, and processes 3-7 each get 6 frequency calculation tasks.

[0134] Each process calculates the frequency according to the assigned frequency calculation task, modifies the corresponding parameters in the environment configuration file (*.env) according to the current sea depth and number of layers, calls the Kraken_mpi.exe executable program to calculate the wave number, and then calculates the complex sound pressure of the sound field at the current frequency. When all frequency calculation tasks of all processes are completed, each process calls the MPI_Gather function to gather all calculation results to process No. 0 to obtain the three-dimensional complex sound pressure, and then uses the Kraken model to calculate the broadband sound propagation loss. The formula is as follows:

[0135]

[0136]

[0137] in, TL B is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; r is the distance dimension; z is the depth dimension; is the broadband sound pressure in a two-dimensional plane;

[0138] Otherwise, when the sea depth at the acquired location is greater than the preset depth, the broadband sound propagation loss is calculated using the Bellhop3D model;

[0139] Bellhop3D is an extension of the Bellhop model in a three-dimensional ocean environment. It uses Gaussian beam tracing theory to calculate the sound field in a horizontal non-uniform environment. The calculation accuracy and reliability of Bellhop3D have been repeatedly tested. It can calculate and obtain a variety of data such as broadband sound propagation loss and sound line propagation trajectory according to specific ocean parameters and sound source configuration; optionally, the Gaussian beam is a geometric Gaussian beam;

[0140] The solution formula of Bellhop3D three-dimensional acoustic ray equation is:

[0141]

[0142] Among them, the propagation trajectory of the sound ray is ( x ( s ), y ( s ), z ( s )), c is the speed of sound.

[0143] The tangent line of the sound ray trajectory is:

[0144]

[0145] The starting position of Bellhop3D sound line ( x (0), y (0), z (0)) is the sound source position ( x s , y s , z s ), the speed of sound at the sound source is c (0), according to the pitch angle α and azimuth β To spread:

[0146]

[0147] Bellhop3D calculates the propagation path of sound rays in the seawater according to the seabed topography, and weights all sound beams in the same direction and adds them up to obtain the sound pressure in that direction. pf

[0148]

[0149] in, u beam is a Gaussian beam, ε 1 、ε 2 is a constant that controls the initial beam width in the two normal directions of the beam;

[0150] In the Bellhop3D calculation process, the previously stored high-resolution terrain data of the global ocean area is used, and the Bellhop3D model is parallelly calculated at the orientation level through the MPI parallel programming technology to shorten the calculation time. Assume that the number of cores available for calculation in the processor is m , indicating that it is executable m Process, broadband sound propagation frequency range a - b Hz, step size c Hz, the broadband ( b - a ) / c +1 frequency of 360 / n The position calculation tasks are evenly distributed to the above processes:

[0151] when m ≥(( b - a ) / c+ 1)×(360 / n ) time, before (( b - a ) / c+ 1)×(360 / n ) processes, each of which gets a computing task;

[0152] When m<(( b - a ) / c+ 1)×(360 / n ) time, before (((( b - a ) / c+ 1)×(360 / n )) mod m ) processes get ((( b - a ) / c+ 1)×(360 / n )) / m +1 computing task, the remaining processes get ((( b - a ) / c+ 1)×(360 / n )) / m A computing task;

[0153] For example, the number of cores available for computing on a processor is 8, the number of processes that can run is 8, and the frequency range of broadband sound propagation is 1001-1005Hz:

[0154] When the step size is 2Hz, there are 3 frequency calculation tasks (1005-1001) / 2+1, which are divided equally at 180° azimuth angle. There are 2 azimuth calculation tasks on each of the 3 frequencies, and a total of 3×2 6 calculation tasks, which is less than the number of processes 8. The 6 processes 0-5 each obtain an azimuth calculation task at a certain frequency, and processes 6-7 are idle.

[0155] When the step size is 1Hz, there are 5 frequency calculation tasks (1005-1001) / 1+1, which are divided into 60° azimuth intervals. There are 6 azimuth calculation tasks on each of the 5 frequencies, and a total of 30 calculation tasks (5×6), which is greater than the number of processes 8. There are 6 processes from 0 to 5, each of which obtains 4 calculation tasks, and processes from 6 to 7 obtain 3 calculation tasks.

[0156] The trajectory of the sound line in the Bellhop3D model is not only related to the pitch angle during propagation. α Related, and no longer fixed to azimuth β , affected by factors such as terrain, horizontal refraction may occur. Each process is opened up in a certain direction calculation task (360 / n ) / m For example, when calculating the azimuth d, the d-th two-dimensional array is used to store the sound pressure obtained by weighted superposition of sound lines that remain at azimuth d during the propagation process, and the remaining two-dimensional arrays store the sound pressure obtained by weighted superposition of sound lines in azimuth d that escape to other azimuths after horizontal refraction. By using Bellhop3D, the horizontal refraction of sound lines under certain conditions can be simulated, which is more in line with the actual ocean sound propagation situation.

[0157] Each process calculates the assigned azimuth task according to the terrain and ocean environment of the current location, modifies the corresponding parameters in the environment configuration file (*.env), calls the Bellhop3D_mpi.exe executable program to calculate the sound line propagation trajectory, and then calculates the complex sound pressure of the sound field at the current azimuth. When all frequency calculation tasks of all processes are completed, each process calls the MPI_Gather function to gather all calculation results to process No. 0 to obtain the three-dimensional complex sound pressure, and then uses the Bellhop3D model to calculate the broadband sound propagation loss. The formula is as follows:

[0158]

[0159]

[0160] in, TL B is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; t is the direction dimension; r is the distance dimension; z is the depth dimension.

[0161] Through testing on an 8-core chip, the parallel speedup ratios of Kraken and Bellhop3D are: the parallel speedup ratio of 8-process Kraken is 7.76 compared to the serial speedup ratio, and the parallel speedup ratio of 8-process Bellhop3D is 7.45 compared to the serial speedup ratio.

[0162] Furthermore, the lightweight dynamic non-cooperative target recognition algorithm is used to determine whether the dynamic non-cooperative target is threatening.

[0163] If the dynamic non-cooperative target is threatening, that is, if the dynamic non-cooperative target is an enemy submarine, manned ship, or medium-to-large offensive unmanned equipment, it is considered threatening. The detection merit is calculated based on whether the target equipment is equipped with active sonar, and the concealment is determined based on the relationship between the detection merit and broadband sound propagation loss.

[0164] When the target device is equipped with active sonar, the active sonar equation is used to calculate the detection figure of merit and determine the magnitude of the detection figure of merit and broadband sound propagation loss.

[0165] Among them, the calculation model of the active sonar equation is:

[0166]

[0167] in, The figure of merit for detection when equipped with active sonar; SL is the active sonar source level;TS The target strength quantitatively describes the strength of the target's reflection ability. It is related to the size and manufacturing material of the unmanned underwater vehicle and is a known parameter. NL The current ocean noise level in the sea area is obtained and recorded by the unmanned underwater vehicle through sonar during navigation; DI For directionality; DT is the detection valve; among them, the active sonar sound source level SL、 Directivity DI and detection threshold DT All are determined by querying existing databases;

[0168] When the detection merit is less than twice the broadband sound propagation loss, the concealment is 0.5, indicating that the evaluation result is that the current concealment is good and not easy to be discovered. Otherwise, that is, when the detection merit is greater than or equal to twice the broadband sound propagation loss, the concealment is 0, indicating that the evaluation result is that the current concealment is poor and is very easy to be discovered. The comparison formula is as follows:

[0169]

[0170] in, The concealment degree when equipped with active sonar;

[0171] When the target device is not equipped with active sonar, the passive sonar equation is used to calculate the detection figure of merit and determine the magnitude of the detection figure of merit and broadband sound propagation loss.

[0172] Among them, the calculation model of the passive sonar equation is:

[0173]

[0174] in, is the detection figure of merit when no active sonar is carried; SL is the noise source level of the radiated noise of the unmanned underwater vehicle, which is related to its own manufacturing and working characteristics and is a known parameter item; NL The current ocean noise level in the sea area is obtained and recorded by the unmanned underwater vehicle through sonar during navigation; DI For directionality; DT For the detection valve; among them, the directional DI and detection threshold DT All are determined by querying existing databases;

[0175] When the detection merit is less than the broadband sound propagation loss, the concealment is 0.5, indicating that the evaluation result is that the current concealment is good and not easy to be discovered. Otherwise, that is, when the detection merit is greater than or equal to the propagation loss, the concealment is 0, indicating that the evaluation result is that the current concealment is poor and is very easy to be discovered. The comparison formula is as follows:

[0176]

[0177] in, This is the concealment degree when no active sonar is installed;

[0178] If the dynamic non-cooperative target is not threatening, that is, the dynamic non-cooperative target is a cruise ship, cargo ship or other vehicle that will not actively attack the unmanned underwater vehicle, then it is not threatening, and the concealment degree is 1, indicating that the evaluation result is that the current concealment is very good and will not be discovered. It can be seen that the detection merit is calculated based on whether the target device is equipped with active sonar, and the current concealment degree of the unmanned underwater vehicle is obtained according to the active and passive sonar equations, providing a basis and support for the implementation of subsequent task formulation, path planning, emergency avoidance and other algorithms.

[0179] refer to Figure 9 As shown, the embodiment of the present invention also provides a method for evaluating the stealth of an unmanned underwater vehicle, comprising the following steps:

[0180] Data acquisition step S1: acquiring hydrological data through sensors carried by the unmanned underwater vehicle, acquiring time domain data through sonar detection of hydroacoustic signals, and obtaining two-dimensional time-frequency data after calculating the time domain data through a time-frequency analysis algorithm;

[0181] Target identification step S2: inputting the two-dimensional time-frequency data into the convolutional neural network model to identify the type of the target device;

[0182] Propagation loss calculation step S3: when the type of the target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained by a positioning device carried by the unmanned underwater vehicle, and the broadband sound propagation loss is calculated in parallel by using the sound propagation model according to the hydrological data;

[0183] Concealment evaluation step S4: calculating the concealment of the unmanned underwater vehicle according to the type of target equipment and the broadband sound propagation loss, and evaluating the concealment of the unmanned underwater vehicle through the concealment;

[0184] Among them, the target recognition step S2 further includes: the compression and expansion module of the convolutional neural network model is configured to include: a compression layer and an expansion layer, the compression layer uses a convolution kernel of a 1×1 convolution kernel to reduce the number of parameters and the number of input channels, and the expansion layer uses a 1×1 convolution kernel and a 3×3 convolution kernel, which are spliced ​​after convolution calculation respectively, and the multi-size convolution kernels perform convolution calculations to retain more feature information;

[0185] A skip connection layer is provided between adjacent compression and expansion modules, for introducing the output of the network layer before the skip connection layer into the output of the network layer after the skip connection layer;

[0186] The convolutional layer of the convolutional neural network model uses depthwise separable convolution to replace the standard 3×3 convolution operation in the compression and expansion module.

[0187] Furthermore, the propagation loss calculation step S3 further includes:

[0188] When the sea depth at the acquired location is less than or equal to the preset depth, the broadband sound propagation loss is calculated using the Kraken model;

[0189] Otherwise, the broadband sound propagation loss is calculated using the Bellhop3D model;

[0190] Among them, the formula for calculating broadband sound propagation loss using the Kraken model is as follows:

[0191]

[0192]

[0193] in, is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; r is the distance dimension; z is the depth dimension; is the broadband sound pressure in a two-dimensional plane;

[0194] Among them, the formula for calculating broadband sound propagation loss through the Bellhop3D model is as follows:

[0195]

[0196]

[0197] in, TL B is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the termination frequency; t is the direction dimension; r is the distance dimension; z is the depth dimension.

[0198] Furthermore, the concealment assessment step S4 further includes:

[0199] If the dynamic non-cooperative target is threatening, its detection figure of merit is calculated based on whether the target equipment is equipped with active sonar, and the concealment degree is determined based on the relationship between the detection figure of merit and broadband sound propagation loss;

[0200] If the dynamic non-cooperative target is not threatening, the concealment is 1;

[0201] When the target device is equipped with active sonar, the active sonar equation is used to calculate the detection figure of merit and determine the magnitude of the detection figure of merit and broadband sound propagation loss.

[0202] When the detection figure of merit is less than twice the broadband sound propagation loss, the concealment is 0.5, otherwise, the concealment is 0;

[0203] When the target device is not equipped with active sonar, the passive sonar equation is used to calculate the detection merit and determine the magnitude of the detection merit and the broadband sound propagation loss.

[0204] When the detection figure of merit is less than the broadband sound propagation loss, the concealment is 0.5, otherwise, the concealment is 0;

[0205] Among them, the calculation model of the active sonar equation is:

[0206]

[0207] in, The figure of merit for detection when equipped with active sonar; SL is the active sonar source level; TS is the target intensity; NL is the ocean noise level; DI For directionality; DT For testing valve;

[0208] Among them, the calculation model of the passive sonar equation is:

[0209]

[0210] in, is the detection figure of merit when no active sonar is carried; SL is the noise source level of the unmanned underwater vehicle radiation; NL is the ocean noise level; DI For directionality; DT For the detection valve.

[0211] It should be noted that the above is a reference method for evaluating the stealth of an unmanned underwater vehicle and a system, and the present invention is not limited thereto.

[0212] The embodiment of the present invention realizes that the dynamic non-cooperative targets in the navigation area can be quickly identified and the broadband sound propagation loss can be quickly calculated through the convolutional neural network model, and finally the concealment assessment level of the environment is obtained, so that it has the ability to quickly and autonomously assess the concealment of the environment, which solves the problem that the existing technology cannot efficiently perform concealment assessment, and can only rely on the signals obtained by passive sonar for automatic judgment without manual assistance, and can only develop a lightweight and fast concealment assessment method based on the characteristics of unmanned underwater vehicles. The technical problem that unmanned underwater vehicles need to rely on automated algorithms to achieve concealment assessment based on lower-level equipment.

[0213] Finally, it should be noted that: the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0214] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present invention, which should be included in the scope of the technical solution for protection of the present invention.

Claims

1. An unmanned underwater vehicle stealth assessment system, characterized in that: include: Data acquisition module: obtains hydrological data through sensors carried by unmanned submersibles, obtains time domain data through sonar detection of hydroacoustic signals, and obtains two-dimensional time-frequency data after calculating the time domain data through time-frequency analysis algorithms; Target recognition module: input the two-dimensional time-frequency data into a convolutional neural network model to identify the type of target device; Propagation loss calculation module: when the type of the target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained by a positioning device carried by the unmanned underwater vehicle, and broadband sound propagation loss is calculated in parallel by a sound propagation model according to the hydrological data; Concealment evaluation module: calculating the concealment of the unmanned underwater vehicle according to the type of the target device and the broadband sound propagation loss, and evaluating the concealment of the unmanned underwater vehicle according to the concealment; The propagation loss calculation module further includes: when the sea depth of the acquired location is less than or equal to a preset depth, broadband sound propagation loss calculation is performed using a Kraken model; otherwise, broadband sound propagation loss calculation is performed using a Bellhop3D model; The Kraken model is parallelized at the frequency level using MPI parallel programming technology. The conditions for allocating frequency calculation tasks are: When m≥(ba) / c+1, the first (ba) / c+1 processes each obtain a frequency calculation task; when m<(ba) / c+1, the first (((ba) / c+1) mod m) processes obtain ((ba) / c+1) / m+1 frequency calculation tasks, and the remaining processes obtain ((ba) / c+1) / m frequency calculation tasks; Wherein, m is the number of processes; ab is the frequency range of broadband sound propagation; c is the step size; The formula for calculating broadband sound propagation loss using the Kraken model is as follows: Wherein, TLB is the broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the ending frequency; r is the distance dimension; z is the depth dimension; pfB is the two-dimensional plane broadband sound pressure; The Bellhop3D model is parallelized at the frequency and orientation levels using MPI parallel programming technology. The conditions for allocating frequency calculation tasks are: When m≥((ba) / c+1)×(360 / n), the first ((ba) / c+1)×(360 / n) processes each obtain a computing task; when m<((ba) / c+1)×(360 / n), the first ((((ba) / c+1)×(360 / n)) mod m) processes obtain (((ba) / c+1)×(360 / n)) / m+1 computing tasks, and the remaining processes obtain (((ba) / c+1)×(360 / n)) / m orientation computing tasks; The formula for calculating broadband sound propagation loss using the Bellhop3D model is as follows: Among them, TLB is broadband sound propagation loss; f is the frequency dimension; a is the starting frequency; b is the ending frequency; t is the direction dimension; r is the distance dimension; z is the depth dimension.

2. The unmanned underwater vehicle stealth assessment system according to claim 1, characterized in that: The concealment assessment module further comprises: If the dynamic non-cooperative target is threatening, the detection figure of merit is calculated according to whether the target device is equipped with an active sonar, and the concealment degree is determined according to the relationship between the detection figure of merit and the broadband sound propagation loss; If the dynamic non-cooperative target is not threatening, the concealment degree is 1.

3. The unmanned underwater vehicle stealth assessment system according to claim 2, characterized in that: The concealment assessment module also includes: When the target device is equipped with an active sonar, the detection figure of merit is calculated using the active sonar equation, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined. When the detection figure of merit is less than twice the broadband acoustic propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0.

4. The unmanned underwater vehicle stealth assessment system according to claim 1, characterized in that: The concealment assessment module also includes: When the target device is not equipped with an active sonar, the passive sonar equation is used to calculate the detection figure of merit, and the magnitude of the detection figure of merit and the broadband sound propagation loss is determined. When the detection figure of merit is less than the broadband acoustic propagation loss, the concealment degree is 0.5, otherwise, the concealment degree is 0.

5. The unmanned underwater vehicle stealth assessment system according to claim 3, characterized in that: The calculation model of the active sonar equation is: in, The figure of merit for detection when equipped with active sonar; SL is the active sonar source level; TS is the target intensity; NL is the ocean noise level; DI For directionality; DT For the detection valve.

6. The unmanned underwater vehicle stealth assessment system according to claim 4, characterized in that: The calculation model of the passive sonar equation is: in, The figure of merit for detection when equipped with passive sonar; SL is the noise source level of the unmanned underwater vehicle radiation; NL is the ocean noise level; DI For directionality; DT For the detection valve.

7. A method for evaluating the stealth of an unmanned underwater vehicle, applicable to the unmanned underwater vehicle stealth evaluation system according to any one of claims 1 to 6, characterized in that: The unmanned underwater vehicle concealment assessment method comprises the following steps: Data acquisition step: acquiring hydrological data through sensors carried by the unmanned underwater vehicle, acquiring time domain data through sonar detection of hydroacoustic signals, and obtaining two-dimensional time-frequency data after calculating the time domain data through a time-frequency analysis algorithm; Target identification step: inputting the two-dimensional time-frequency data into a convolutional neural network model to identify the type of target device; Propagation loss calculation step: when the type of the target device is identified as a dynamic non-cooperative target, the position of the unmanned underwater vehicle is obtained by a positioning device carried by the unmanned underwater vehicle, and broadband sound propagation loss is calculated in parallel by a sound propagation model according to the hydrological data; Concealment evaluation step: Calculate the concealment of the unmanned underwater vehicle according to the type of the target device and the broadband sound propagation loss, and evaluate the concealment of the unmanned underwater vehicle through the concealment.

Images