Underwater acoustic communication signal passive fusion detection method and device based on multi-node multi-scale feature fusion

Abstract

The invention provides an underwater acoustic communication signal passive fusion detection method and device based on multi-node multi-scale feature fusion. The method comprises the steps that a plurality of hydrophone nodes are arranged at a plurality of spatial positions of a monitored area, the hydrophone nodes are used for receiving underwater acoustic communication signals, the received underwater acoustic communication signals are preprocessed, preprocessing comprises noise reduction and short-time Fourier transform (STFT) on the underwater acoustic communication signals after noise reduction, and the underwater acoustic communication signals are sent to the monitored area. Obtaining a plurality of corresponding time-frequency diagrams; inputting the plurality of time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features; and inputting the fused features into a classifier to detect whether a communication signal exists in the fused features. According to the invention, the information utilization degree of the received signal can be improved, and the detection rate under a low signal-to-noise ratio is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of passive detection of underwater acoustic signals, and in particular to a method and device for passive fusion detection of underwater acoustic communication signals based on multi-node multi-scale feature fusion. Background Art

[0002] Passive detection of underwater acoustic signals refers to the process of judging whether there is a target signal of interest from the passively received underwater acoustic signals. It is an important research topic in the field of marine acoustic monitoring. According to their sound sources and waveform characteristics, underwater acoustic signals are mainly divided into marine environmental noise, marine animal calls, ship radiation noise, active sonar signals and underwater acoustic communication (UWAC) signals. Among them, UWAC signals are the carriers of underwater information transmission, and their signal waveforms carry the transmitted information. Therefore, passive detection of UWAC signals is of great significance to improving the efficiency of underwater acoustic communication networks and underwater acoustic perception capabilities.

[0003] Since the underwater acoustic channel has strong spatial variability, the signal transmission path is related to factors such as the sound velocity profile and ocean geography. The signal strength and quality of the receiving hydrophone at different depths and different directions vary greatly. In order to improve the signal perception capability, it is necessary to deploy multiple nodes to receive signals at multiple spatial locations, and overcome the problems of inability to receive, incomplete reception, and inaccurate judgment caused by the spatial differences in sound wave propagation through multi-node signal fusion perception. Signal detection based on multi-node information fusion can process data and information from multiple information sources from multiple angles and levels, thereby improving the accuracy of state and identity estimation, and can solve this problem to a certain extent, thereby improving the accuracy and reliability of UWAC signal detection.

[0004] In the fusion process, traditional methods mainly use the Neyman-Pearson criterion (NP criterion) and the minimum risk Bayes criterion to fuse the detection values ​​of each sensor node to form a fusion likelihood ratio (LRT). The main fusion methods are fusion at the decision level and fusion at the feature level. The decision-level fusion uses single-bit quantization, and each node forms a hard decision. Finally, the decision-level fusion is realized in the fusion center, and the fusion data loss is the largest. In the research of feature-level fusion, there are methods with multiple quantization, and the fusion performance is improved compared with the decision-level fusion. However, most of the above traditional signal fusion detection methods are based on domain knowledge, construct test statistics, and need to set corresponding thresholds according to different environments. The performance is seriously degraded under low signal-to-noise ratio conditions, and some methods also require certain prior knowledge of signals and noise, which is difficult to work effectively under third-party reception conditions. In order to extract more effective and robust classification features and improve the robustness to the environment, in recent years, some literatures have proposed some methods based on deep learning, which automatically complete the feature extraction and classification links in traditional methods by deep neural networks.

[0005] The main target signal of the research on using deep learning to achieve passive signal fusion detection is radio signals. Currently, there is no deep learning method to achieve multi-node passive fusion detection of UWAC signals. Moreover, these passive fusion detection methods for radio signals mainly achieve multi-node fusion at the decision level to achieve the final signal detection; similarly, the modulation recognition network MHFNet that currently uses features to achieve fusion loses a lot of information during feature extraction and fusion, resulting in low signal detection accuracy. Research on feature extraction and fusion at the feature level can improve the information utilization of the received signal and is of great value in improving the accuracy of signal detection. Summary of the invention

[0006] In view of the problem that the existing passive fusion detection methods lose too much useful information during decision-level fusion and have low detection rate under low signal-to-noise ratio, the present invention proposes a UWAC signal passive fusion detection method and device based on multi-node multi-scale feature fusion. By extracting multi-scale features of multiple nodes and fusing the features, full utilization of multi-node received signal information is achieved, and finally passive fusion detection of UWAC signals is realized.

[0007] In a first aspect, the present invention provides a passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion, wherein multiple hydrophone nodes are deployed at multiple spatial positions in a monitored area, and multiple hydrophone nodes are used to receive underwater acoustic communication signals, including:

[0008] Step 1: preprocessing the received multiple underwater acoustic communication signals, wherein the preprocessing includes noise reduction and performing short-time Fourier transform (STFT) on the multiple underwater acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams;

[0009] Step 2: Input multiple time-frequency graphs into the preset multi-node multi-scale feature fusion network to obtain the fused features;

[0010] Step 3: Input the fused features into the classifier to detect whether there is a communication signal in the fused features.

[0011] Furthermore, the multi-node multi-scale feature fusion includes a feature extraction module and a feature fusion module; wherein the feature extraction module includes multiple basic branches corresponding to multiple time-frequency graphs, each basic branch includes a feature extraction unit and a feature reconstruction unit, the feature extraction unit is used to extract the multi-scale features of the corresponding time-frequency graph, and the feature reconstruction unit is used to reconstruct the size dimensions of the multi-scale features to be consistent; the feature fusion module is used to first fuse the multi-scale features of a single time-frequency graph to obtain the features of a single time-frequency graph, and then fuse the features of multiple time-frequency graphs to obtain the fused features.

[0012] Furthermore, the feature extraction unit adopts an improved ResNet network; the improved ResNet network refers to a network formed after removing the average pooling layer and the fully connected layer in the ResNet network.

[0013] Furthermore, the feature reconstruction unit includes a convolutional layer and a transposed convolutional layer connected in sequence.

[0014] Furthermore, the feature fusion module includes a group concatenation G-concat layer and a group convolution G-conv layer;

[0015] Correspondingly, the multi-scale features of a single time-frequency graph are fused to obtain the features of a single time-frequency graph, specifically including: using the G-concat layer to fuse the features of a single time-frequency graph according to the following formula x i :

[0016]

[0017] Where i is the number of the hydrophone node, and Represents the two highest-level feature vectors in the multi-scale features of the time-frequency graph corresponding to the hydrophone node i, m i Indicates that the multi-scale features and Features of other scales other than ;

[0018] The features of multiple time-frequency graphs are fused to obtain fused features, specifically including: using the grouped convolution G-conv layer to perform two grouped convolutions in sequence according to the following formula to obtain the fused feature U = {u1, u2};

[0019]

[0020] Among them, W and V represent the intermediate fusion features of the first group convolution output. and represents the two different convolution kernels corresponding to the hydrophone node i in the first group convolution process; and Represents two different convolution kernels in the second group convolution process.

[0021] Furthermore, the classifier adopts a CNN-based binary classification network.

[0022] Furthermore, it also includes: pre-training a preset multi-node multi-scale feature fusion network and a classifier using a migration data training set; wherein the migration data training set refers to a data set composed of data having the same distribution as the transmitted signal in the test data.

[0023] In a second aspect, the present invention provides a passive fusion detection device for underwater acoustic communication signals based on multi-node multi-scale feature fusion, comprising:

[0024] A plurality of hydrophone nodes are arranged at a plurality of spatial locations in the monitored area for receiving hydroacoustic communication signals;

[0025] A preprocessing module is used to preprocess the received multiple underwater acoustic communication signals, wherein the preprocessing includes noise reduction and short-time Fourier transform (STFT) of the multiple underwater acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams;

[0026] A fusion network module is used to input multiple time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features;

[0027] The classification module is used to input the fused features into the classifier to detect whether there is a communication signal in the fused features.

[0028] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0029] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0030] Beneficial effects of the present invention:

[0031] (1) The present invention uses the time-frequency diagram as the input of the fusion network. Since the time-frequency diagram depicts the energy distribution of the signal at different times and frequencies, it has the advantages of both time domain and frequency domain representation and contains rich features. Therefore, by extracting and fusing the features of the time-frequency diagram at the feature level, the signal can be detected, which can improve the information utilization of the received signal and enhance the detection rate under low signal-to-noise ratio.

[0032] (2) The present invention proposes a signal processing flow that can be used for multi-node fusion passive detection of UWAC signals under low signal-to-noise ratio conditions. That is, the detection of UWAC signals is achieved through preprocessing, MMFFN network and finally CNN-based binary classification network. The proposed method can achieve a higher detection rate under low signal-to-noise ratio conditions.

[0033] (3) The present invention designs a multi-node multi-scale feature fusion deep learning network MMFFN, which can realize multi-scale extraction and fusion of features of signals received by multiple receiving sources, and fully utilize the information of multi-node received signals. Feature-Extraction and Feature-Fusion modules are innovatively designed in this network; among them, the Feature-Extraction module innovatively uses the ResNet network as the backbone network for multi-scale feature extraction and forms BasicNet, so that deeper features can be extracted without the problem of gradient explosion. In order to realize the fusion of multi-node multi-scale features, it is proposed in the Feature-Fusion module to group and step by step realize the fusion of multi-node and multi-scale features by means of group connection and group convolution.

[0034] (4) The experimental results show that when the number of nodes is 3, the present invention can s When / N0=-2dB, the detection rate reaches 98%. It can effectively detect common UWAC signals in unknown channels, including multi-level frequency shift keying (MFSK), multi-level phase shift keying (MPSK), orthogonal frequency division multiplexing (OFDM) and linear frequency modulation (LFM) signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic flow chart of a method for passive fusion detection of underwater acoustic communication signals based on multi-node multi-scale feature fusion provided by an embodiment of the present invention;

[0036] Figure 2 An architecture diagram of a passive fusion detection model for underwater acoustic communication signals based on multi-node multi-scale feature fusion provided by an embodiment of the present invention;

[0037] Figure 3 The improved ResNet network provided by the embodiment of the present invention;

[0038] Figure 4 BasicNet, a basic branch provided by the embodiment of the present invention;

[0039] Figure 5 The group concatenation G-concat layer provided by the embodiment of the present invention;

[0040] Figure 6 The grouped convolution G-conv layer provided by the embodiment of the present invention;

[0041] Figure 7 A classifier provided by an embodiment of the present invention;

[0042] Figure 8 The test results of different numbers of nodes provided by the embodiments of the present invention;

[0043] Fig. 9 The impulse responses of different channels provided by the embodiments of the present invention;

[0044] Fig.10 The test results of different channels provided by the embodiment of the present invention;

[0045] Fig.11 A schematic diagram of the structure of a passive fusion detection device for underwater acoustic communication signals based on multi-node multi-scale feature fusion provided by an embodiment of the present invention;

[0046] Fig.12 A structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all 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.

[0048] The application background of the solution of the present invention is to deploy multiple hydrophone nodes (hereinafter referred to as nodes) at multiple spatial locations in the monitored area, and use multiple hydrophone nodes to receive water acoustic communication signals. In order to better understand the technical solution of the present invention, based on the above application background, the technical problem to be solved by the present invention is first modeled as follows.

[0049] In the case of multi-node reception, the UWAC signal received by the i-th node can be expressed as:

[0050]

[0051] In the formula, r i (t) represents the received signal of the i-th node, x(t) represents the target signal, and h i (t) represents the impulse response of the underwater acoustic channel when the target signal reaches the i-th node, w i (t) is the ocean ambient noise at the ith node, N node is the number of nodes.

[0052] The present invention assumes that there is no relative displacement between the target sound source and the receiving node, so the underwater acoustic channel can be modeled as a coherent multipath channel, and its model can be expressed as:

[0053]

[0054] Where δ(t) is the impulse function, K i is the total number of sound rays reaching the i-th node, i = 1, 2, ..., N node , A m is the amplitude of the mth sound ray, τ m is the transmission delay of the mth sound ray.

[0055] For UWAC signals, the symbol signal-to-noise ratio is usually used to measure the noise level of the received signal. It is defined as the ratio of the energy of each symbol to the noise power spectrum density, denoted by E s / N0; where E s represents the signal energy of each symbol, and N0 represents the noise power spectral density.

[0056] Therefore, after sampling the model, it can be expressed as:

[0057]

[0058] Among them, r i (n), x(n), h i (n) and w i (n) respectively by r i (t), x(t), h i (t) and wi (t) with sampling rate f s Sampling obtained.

[0059] Based on the above content, the present invention performs passive fusion detection of UWAC signals on multiple nodes by node The signal r received by the nodes in the same time period i (n),i=1,2,…,N node The process of joint processing to determine whether there is a target communication signal. Therefore, it can be described as a binary hypothesis testing problem:

[0060]

[0061] Among them, H1 is defined as the presence of target signal, and H0 is defined as the absence of target signal. In the detection model, the false alarm probability (P f ) and detection probability (P d ) to evaluate the performance of the signal detection algorithm:

[0062] P f =P(H1|H0)(5)

[0063] P d =P(H1|H1)(6)

[0064] In order to solve the above technical problems, the present invention proposes a UWAC signal passive fusion detection method based on multi-node multi-scale feature fusion, such as Figure 1 As shown, the following steps are included:

[0065] S101: pre-processing a plurality of received underwater acoustic communication signals;

[0066] Specifically, since UWAC signals are usually distributed above 2kHz, the signal preprocessing stage will first perform noise reduction preprocessing on the multi-node sampling received signals, including: filtering out the noise below 2kHz and the radiation noise of the ship through frequency domain filtering.

[0067] Next, the multiple underwater acoustic communication signals after noise reduction are subjected to short-time Fourier transform (STFT) to obtain the corresponding multiple time-frequency diagrams. Specifically, the noise-reduced signals are processed in segments, with 1s as the time window, and the data r(n) in the window is normalized and then subjected to STFT, which is expressed as follows:

[0068]

[0069] Where N is the number of sampling points, is the window function, here it is the Hanning window, the window length is 300, the step length is 30, and the time-frequency diagram T of fixed output size is obtained. m .

[0070] S102: Inputting multiple time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features;

[0071] S103: Input the fused features into a classifier to detect whether there is a communication signal in the fused features.

[0072] Specifically, the time-frequency diagram depicts the energy distribution of the signal at different times and frequencies. It has the advantages of both time and frequency domain representation and contains rich features. Therefore, the embodiment of the present invention uses the time-frequency diagram as the input of the fusion network, extracts and fuses the features of the time-frequency diagram at the feature level to realize signal detection, which can improve the information utilization of the received signal and improve the detection rate under low signal-to-noise ratio.

[0073] In one embodiment, Figure 2 As shown, an embodiment of the present invention provides a passive fusion detection model for underwater acoustic communication signals, and the network model includes a signal preprocessor (Preprocessor), a multi-node multi-scale feature fusion network (Multi-node multi-scale feature fusion network, MMFFN) and a binary classifier (Classifier). The detection process of the network model includes: first, the signal preprocessor performs noise reduction preprocessing on the multi-node sampled received signal and performs STFT time-frequency transformation, and then the obtained multi-node time-frequency diagram is used as input and sent to the MMFFN network to realize the extraction and fusion of multi-node multi-scale features, and finally the fused features are sent to the binary classifier to realize further extraction and classification of features, thereby completing the detection of the existence of the signal. The workflow of the signal preprocessor is the same as that of the above embodiment, and will not be repeated here. The following is a detailed introduction to MMFFN.

[0074] Multi-scale feature extraction and fusion is an important concept in the field of computer vision, especially in tasks such as target detection and image recognition. This technology can better capture the various manifestations of the target by combining features of different scales, thereby improving the accuracy and stability of detection. The MMFFN proposed in the embodiment of the present invention is an end-to-end trainable convolutional neural network that can extract and fuse multi-node multi-scale features. The network as a whole consists of two parts, namely a feature extraction (Feature-Extraction) module and a feature fusion (Feature-Fusion) module. Among them, Feature-Extraction includes multiple basic branches BasicNet, one basic branch corresponds to the time-frequency diagram of a node; each basic branch mainly completes the extraction of multi-scale features of the node; the Feature-Fusion module completes the connection and fusion of multi-node multi-scale features.

[0075] In an embodiment of the present invention, each basic branch includes a feature extraction unit and a feature reconstruction unit. The feature extraction unit is used to extract multi-scale features of the corresponding time-frequency graph, and the feature reconstruction unit is used to reconstruct the size dimensions of the multi-scale features to be consistent. As an implementable method, the feature extraction unit adopts an improved ResNet network; the improved ResNet network refers to a network formed after removing the average pooling layer and the fully connected layer in the ResNet network. This is because ResNet introduces residual connections, which allow input signals to be directly transmitted to deeper layers of the network, allowing ResNet to train a very deep network structure without performance degradation. However, when extracting multi-scale features, the present invention requires features extracted by a deeper network. Therefore, the present invention chooses to modify the original ResNet and then use it as the backbone network. The structure of the modified ResNet is as follows: Figure 3 To extract multi-scale features, the embodiment of the present invention removes the average pooling layer and the fully connected layer in the original ResNet, and retains the bottom convolutional block and the four-layer bottleneck structure. Figure 4 As shown in Figure 1, a single time-frequency graph is input into the backbone network as input, and the convolution features at five scales from R-1 to R-5 are obtained, as shown in Figure 1. Figure 4 The medium blue squares are the stack of features.

[0076] Since the features obtained at the five scales cannot be directly fused, the present invention also designs a feature reconstruction module in the basic branch to achieve upsampling through convolution and transposed convolution, and reconstruct the feature size dimensions to be consistent. The structure is as follows: Figure 4In the reconstruction, since the top high-level features of R-5 are encoded with high-dimensional information and have a rich receptive field, this feature can be used to achieve the role of classification supervision at the top. The embodiment of the present invention outputs activation vectors of C categories (C = 2 at this time) The bottom-level features from R-1 to R-4 are beneficial to enhancing the top-level classification, and the corresponding output single-channel feature map is

[0077]

[0078] In the embodiment of the present invention, the Feature Fusion module includes two parts: a group concatenation (G-concat) layer and a group convolution (G-conv) layer. The G-concat layer completes the connection combination of multi-node multi-scale features, and the G-conv layer implements supervised feature fusion. The process of the G-concat layer is as follows: Figure 5 As shown in the figure, num represents the number of nodes. m After BasicNet, we get the activation vector A i And the single channel feature map m i When the number of nodes is N node When and In order to enable the activation vector to achieve supervision, we copy the underlying feature maps from R-1 to R-4 The two top-level activation vectors are connected in groups to obtain a combination of feature maps and activation vectors. For a single time-frequency graph, the above combination process can be expressed as follows:

[0079]

[0080] in, Composed of A i After obtaining the combination of feature maps and activation vectors, we implement supervised feature fusion through group convolution, and its input and output are as follows: Figure 6 As shown in the figure, num represents the number of nodes. Group convolution can increase the diagonal correlation between filters in adjacent layers, thus helping the model learn more discriminative features. The specific implementation is as follows Figure 6 shown.

[0081] In the first group convolution, we As a group, convolution is performed to achieve multi-scale fusion of each node. The number of input channels of the first group convolution is as shown in formula (9).

[0082]

[0083] At this time, the length of F is 5*C*N node , where 5 represents the 5-layer feature of R-(1-5), C represents the classification category (C = 2 in this case), and N node Indicates the number of nodes. The number of output channels of the first grouped convolution is C*N node , the number of groups is N node , then the number of input channels per group is 5*2, and the number of output channels per group is 2. We define the different convolution kernels in the first group convolution process as Then the output after the first group convolution is It can be expressed as:

[0084]

[0085] In the second group convolution, we implement convolution by treating W and V as a group. The number of input channels of the second group convolution is 2*N. node , the number of output channels is 2, the number of groups is 2, then the number of input channels per group is N node , the number of output channels in each group is 1. The two convolution kernels of the second group convolution are Then the output U = {u1,u2} after the second group convolution can be expressed as:

[0086]

[0087] In an embodiment of the present invention, the Feature-Extraction module innovatively uses the ResNet network as the backbone network for multi-scale feature extraction and forms a BasicNet, so that deeper features can be extracted without the problem of gradient explosion; in the Feature-Fusion module, it is proposed to group and fuse multi-node and multi-scale features in steps by means of group connection and group convolution; therefore, the MMFFN provided by the present invention can realize multi-scale extraction and fusion of features of signals received by multiple receiving sources, and fully utilize the information of multi-node received signals.

[0088] On the basis of the above embodiments, after obtaining the fused features, a CNN-based binary classification network C is constructed in the embodiment of the present invention as a classifier to automatically extract the detection features used to characterize the presence or absence of the signal and complete the binary classification recognition. The network structure is as follows Figure 7The working process of the classifier includes: first, the fused features are input into the binary classification module; then two convolution blocks, two maximum pooling layers, two fully connected layers and Softmax activation function are used to extract and classify high-dimensional features; the probability of the presence of a communication signal and pure noise in the received signal is output respectively, and the activation function is set in the same way as the decision device. At this time, the predicted label probability vector output by the Softmax classifier is It can be expressed as:

[0089]

[0090] S=f CNN (U)(13)

[0091] Among them, L p represents the predicted label value, S represents the input vector of Softmax, and f CNN Represents the nonlinear function formed by the network before the Softmax layer in classifier C.

[0092] It is understandable that the above MMFFN and classifier need to be trained first, and only after training can they have the fusion and classification capabilities respectively. Therefore, based on the above embodiments, the embodiment of the present invention provides the following network training method.

[0093] (1) Constructing a migration training dataset

[0094] The core of the network training process is to find the mapping to the feature space based on the sampled training data. In order to make the network trained under the training set still effectively realize this mapping on the test data, it is usually expected that the offline training data set and the online test data set have the same distribution. However, in the application of UWAC signal detection, when facing unfamiliar waters and different communication scenarios, the information of the target channel is difficult to obtain accurately, so it is difficult to obtain a sufficient number of training data with the same distribution as the test data. Therefore, the embodiment of the present invention utilizes the idea of ​​transfer learning to complete the training of the network by constructing a data model containing the same distribution as the signal sent in the test data to generate a migration data training set. The migration data model is:

[0095]

[0096] in, represents the received signal, x(n) is the transmitted signal, and the signal type is the same as that in equation (3). It's noise.

[0097] Compared with equation (3), equation (14) simplifies the influence of underwater acoustic multipath channels. Although the distribution of data generated by the two models has changed, they both contain the same set of transmitted signals, i.e. The waveforms of and r(n) contain the same useful signal components, so they are transferable. The modulation parameters of typical UWAC signals and the working parameters of existing experimental equipment in the laboratory are comprehensively considered, and the signal is generated based on the model described by formulas (1) to (3). Detailed signal parameters are shown in Table 1, where " / " means that this parameter does not need to be set, and "{}" means random selection within the given set. The signal sampling rate is uniformly set to 48kHz, the length is 8192 sampling points, about 171ms, and at the same time, the center frequency of the LFM signal and the carrier frequency of other signals are set to 15-16kHz, and the OFDM subcarrier is BPSK or QPSK.

[0098] Table 1 Signal parameters

[0099]

[0100] During the training phase, each hydrophone generates 30 samples for each type of signal under different signal-to-noise ratio conditions. In order to make the number of signals the same as the number of pure noise, the number of pure noise samples generated is 30*6=180. In the case of a single node, we set the signal-to-noise ratio range of each node symbol to [-6,20]dB, with a step size of 2dB. At this point, we get a total of 5040 samples. In the case of three nodes, we set the signal-to-noise ratio range of each node symbol to [-6,14]dB, and the step sizes of the three nodes are 4, 6, and 8dB respectively, for a total of 25920 samples.

[0101] (2) Model training

[0102] In such Figure 2 In the signal detection model shown in Figure 1, effective signal detection is achieved by jointly training the MMFFN and the binary classification network. The cross entropy loss function is used during training, which can be expressed as:

[0103]

[0104] Among them, S is the parameter of the network model, N b is the number of samples in a batch. When the loss function converges, the network training is completed. The training learning rate of the model is 0.001, and the batch size is N. b =32, the SGD optimizer (momentum=0.9, weight_decay=0.0001) was used to optimize the loss function, and all data were iterated until the overall loss function of the model converged and then the training was stopped.

[0105] In a multi-hydrophone receiving system, due to the receiving position, environment, and distance, the transmitted signal reaches the receiving end through different channels. In order to be close to the actual application scenario, we selected two different water areas as the water environment based on the Argo ocean database, recorded as Water Area 1 and Water Area 2. Water Area 1 is a typical deep-sea acoustic profile, which contains typical ocean acoustic characteristics. Water Area 2 is a typical shallow-sea acoustic profile. In order to reflect the characteristics of the water acoustic channel as fully as possible, the sound source positions are selected in the transition layer, main transition layer, and sound channel axis. Then, the Bellhop water acoustic channel simulation software is used to generate linear time-invariant water acoustic channels under different transmission conditions. Its parameters are shown in Table 2.

[0106] Table 2 Underwater acoustic channel parameters

[0107]

[0108] The transfer function of each channel is as follows:

[0109] H1(z)=0.04+z -353 +0.508z -570 +0.283z -644 (16)

[0110] H2(z)=0.32+0.45z -48 +z -61 +0.9318z -267 (17)

[0111] H3(z)=0.68+z -184 +0.882z -403 (18)

[0112] H4(z)=0.2039+0.3178z -2327 +0.0759z -2938 +z -3071 (19)

[0113] +0.5875z -3711 +0.8845z -4470 +0.3302z -11697

[0114] In the test phase, we also randomly select channels to generate signal samples according to the number of hydrophones according to equation (14) to construct the test set. We select different signal-to-noise ratio combinations according to the purpose of the test, and generate 300 signal samples and 300 pure noise samples at each signal-to-noise ratio, that is, a total of 600 test samples at each signal-to-noise ratio.

[0115] In order to verify the effectiveness of the solution of the present invention, the present invention also provides the following effectiveness verification experimental data.

[0116] (1) Verification of the effectiveness of multi-scale feature fusion

[0117] The effectiveness of the method of the present invention is illustrated by comparing the detection performance differences between the method of the present invention and several deep learning methods under Gaussian noise conditions. The results are shown in Table 3. The false alarm rate of the method of the present invention is 0.33%, and the detection rate reaches 98.33% under SNR = -15dB. It can be seen that the method of the present invention performs better in both detection rate and false alarm rate. In fact, most of the existing detection methods based on deep learning directly extract features from the original speech signal, while the method of the present invention achieves better results by extracting multi-scale features for detection.

[0118] Table 3 Comparison of detection performance of different methods

[0119]

[0120] Document 1 in Table 3 is "Li Y, Wang B, Shao G, et al. Blind Detection of UnderwaterAcoustic Communication Signals Based on Deep Learning [J]. IEEE Access, 2020,8:204114-204131"; Document 2 is "Ke D, Huang Z, Wang X, et al. Blind Detection Techniques for Non-Cooperative Communication Signals Based on Deep Learning Learning[J].IEEEAccess,2019,7:89218-89225."; Document 3 is "Gao J, Yi X, Zhong C, et al.Deep Learning for Spectrum Sensing[J].IEEE Wireless Communications Letters, 2019,8(6):1727-1730".

[0121] (2) Verification of the effectiveness of multi-node and multi-scale feature fusion

[0122] In order to verify the effectiveness of multi-node multi-scale feature fusion for UWAC signal detection, we realize signal detection by increasing the number of nodes and compare it with the detection performance of a single node. At this time, in order to illustrate the detection gain brought by multi-node feature passive fusion detection, the multi-node test set is set to make the signal-to-noise ratio of each node equal, and randomly select channels according to the number of hydrophones, with 600 test samples at each signal-to-noise ratio. Through such a setting, the detection gain brought by the increase of the signal-to-noise ratio of a certain node when multi-nodes are tested is avoided, thereby strengthening the detection gain brought by multi-node feature passive fusion detection. The trained single-node, three-node and five-node network models are tested in test sets with different numbers of nodes, and the detection results are shown as follows. Figure 8 .

[0123] As can be seen from the figure, the passive fusion detection of multiple nodes has improved the detection rate compared with a single node, and the detection signal-to-noise ratio gain can reach about 3dB; and at a lower signal-to-noise ratio, multiple nodes have a higher detection rate. s When / N0=-8dB, the detection rate of a single node is 20%, while that of multiple nodes reaches 46% to 48%. The detection rates of three nodes and five nodes are similar. At the same time, the false alarm rate of five nodes is 0.13%, while that of three nodes is 1.4%. Five nodes perform better in terms of false alarm rate. This shows that increasing the number of nodes can bring better detection gain.

[0124] (3) Algorithm reliability verification

[0125] To verify the reliability of the algorithm, we conducted experiments in different underwater acoustic channels to verify the robustness of the method of the present invention for different underwater acoustic multipath channels. To ensure the diversity of channel characteristics, we adopted the above underwater acoustic channel parameters. At a sampling rate of 48kHz, the maximum propagation delays of the four underwater acoustic channels relative to their direct paths are 18.25ms, 26.27ms, 74.60ms, 6.48ms and 87.77ms respectively. Fig. 9 The impulse response of each channel is given in . The experimental results are shown in Fig.10 As shown in Figure 2, it can be seen that the detection performance under different underwater acoustic channel conditions is close, with little difference. This result proves the robustness of this method for different underwater acoustic multipath channels.

[0126] like Fig.11 As shown, an embodiment of the present invention provides a passive fusion detection device for underwater acoustic communication signals based on multi-node multi-scale feature fusion, including: multiple hydrophone nodes, a preprocessing module, a fusion network module and a classification module.

[0127] Among them, multiple hydrophone nodes are deployed at multiple spatial positions in the monitored area to receive water acoustic communication signals; the preprocessing module is used to preprocess the multiple received water acoustic communication signals, and the preprocessing includes noise reduction and short-time Fourier transform STFT of the multiple water acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency graphs; the fusion network module is used to input the multiple time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features; the classification module is used to input the fused features into a classifier to detect whether there is a communication signal in the fused features.

[0128] It should be noted that the passive fusion detection device for underwater acoustic communication signals based on multi-node multi-scale feature fusion provided in the embodiment of the present invention is for the purpose of realizing the above-mentioned method. Its specific functions can be referred to the above-mentioned method embodiments and will not be repeated here.

[0129] Fig.12 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.12 As shown, the electronic device may include: a processor 1201, a communication interface 1202, a memory 1203 and a communication bus 1204, wherein the processor 1201, the communication interface 1202 and the memory 1203 communicate with each other through the communication bus 1204. The processor 1201 may call the logic instructions in the memory 1203 to execute a passive fusion detection method for underwater acoustic communication signals, the method comprising: placing multiple hydrophone nodes at multiple spatial positions in the monitored area, and using multiple hydrophone nodes to receive underwater acoustic communication signals; preprocessing the received multiple underwater acoustic communication signals, the preprocessing comprising denoising and performing short-time Fourier transform (STFT) on the multiple underwater acoustic communication signals after denoising to obtain multiple corresponding time-frequency graphs; inputting the multiple time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features; and inputting the fused features into a classifier to detect whether there is a communication signal in the fused features.

[0130] In addition, when the logic instructions in the above-mentioned memory 1203 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0131] An embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: deploying multiple hydrophone nodes at multiple spatial positions in the monitored area, and using multiple hydrophone nodes to receive water acoustic communication signals; preprocessing the received multiple water acoustic communication signals, wherein the preprocessing includes noise reduction and short-time Fourier transform (STFT) of the multiple water acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams; inputting the multiple time-frequency diagrams into a preset multi-node multi-scale feature fusion network to obtain fused features; and inputting the fused features into a classifier to detect whether there are communication signals in the fused features.

[0132] An embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the methods provided by the above-mentioned method embodiments are implemented, for example, including: deploying multiple hydrophone nodes at multiple spatial positions in the monitored area, and using multiple hydrophone nodes to receive water acoustic communication signals; preprocessing the multiple received water acoustic communication signals, wherein the preprocessing includes noise reduction and short-time Fourier transform (STFT) of the multiple water acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams; inputting the multiple time-frequency diagrams into a preset multi-node multi-scale feature fusion network to obtain fused features; and inputting the fused features into a classifier to detect whether there is a communication signal in the fused features.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion, wherein multiple hydrophone nodes are deployed at multiple spatial locations in the monitored area, and multiple hydrophone nodes are used to receive underwater acoustic communication signals, characterized in that: include: Step 1: preprocessing the received multiple underwater acoustic communication signals, wherein the preprocessing includes noise reduction and performing short-time Fourier transform (STFT) on the multiple underwater acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams; Step 2: Input multiple time-frequency graphs into the preset multi-node multi-scale feature fusion network to obtain the fused features; Step 3: Input the fused features into the classifier to detect whether there is a communication signal in the fused features.

2. According to claim 1, a passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion is characterized in that: The multi-node multi-scale feature fusion includes a feature extraction module and a feature fusion module; wherein the feature extraction module includes a plurality of basic branches corresponding to a plurality of time-frequency graphs, each basic branch includes a feature extraction unit and a feature reconstruction unit, the feature extraction unit is used to extract the multi-scale features of the corresponding time-frequency graph, and the feature reconstruction unit is used to reconstruct the size dimensions of the multi-scale features to be consistent; the feature fusion module is used to first fuse the multi-scale features of a single time-frequency graph to obtain the features of a single time-frequency graph, and then fuse the features of multiple time-frequency graphs to obtain the fused features.

3. According to claim 2, a passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion is characterized in that: The feature extraction unit adopts an improved ResNet network; the improved ResNet network refers to a network formed after removing the average pooling layer and the fully connected layer in the ResNet network.

4. According to claim 2, a passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion is characterized in that: The feature reconstruction unit includes a convolutional layer and a transposed convolutional layer connected in sequence.

5. The passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion according to any one of claims 1 to 4, characterized in that: The feature fusion module includes a group concatenation G-concat layer and a group convolution G-conv layer; Correspondingly, the multi-scale features of a single time-frequency graph are fused to obtain the features of a single time-frequency graph, specifically including: using the G-concat layer to fuse the features of a single time-frequency graph according to the following formula x i : Where i is the number of the hydrophone node, and Represents the two highest-level feature vectors in the multi-scale features of the time-frequency graph corresponding to the hydrophone node i, m i Indicates that the multi-scale features and Features of other scales other than ; The features of multiple time-frequency graphs are fused to obtain fused features, specifically including: using the grouped convolution G-conv layer to perform two grouped convolutions in sequence according to the following formula to obtain the fused feature U = {u1, u2}; Among them, W and V represent the intermediate fusion features of the first group convolution output. and represents the two different convolution kernels corresponding to the hydrophone node i in the first group convolution process; and Represents two different convolution kernels in the second group convolution process.

6. The passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion according to any one of claims 1 to 4, characterized in that: The classifier adopts a CNN-based binary classification network.

7. The passive fusion detection method for underwater acoustic communication signals based on multi-node multi-scale feature fusion according to any one of claims 1 to 4, characterized in that: Also includes: The migration data training set is used in advance to jointly train the preset multi-node multi-scale feature fusion network and classifier; The migration data training set refers to a data set consisting of data having the same distribution as the transmitted signals in the test data.

8. A passive fusion detection device for underwater acoustic communication signals based on multi-node multi-scale feature fusion, characterized in that: include: A plurality of hydrophone nodes are arranged at a plurality of spatial locations in the monitored area for receiving hydroacoustic communication signals; A preprocessing module is used to preprocess the received multiple underwater acoustic communication signals, wherein the preprocessing includes noise reduction and short-time Fourier transform (STFT) of the multiple underwater acoustic communication signals after noise reduction to obtain corresponding multiple time-frequency diagrams; A fusion network module is used to input multiple time-frequency graphs into a preset multi-node multi-scale feature fusion network to obtain fused features; The classification module is used to input the fused features into the classifier to detect whether there is a communication signal in the fused features.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images