Underwater acoustic multi-harmonic micro-doppler feature recognition technology based on CAMobileNetV2

Abstract

This invention belongs to the field of sonar signal processing and proposes an underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2. This technology establishes a micro-echo model of underwater vehicle propeller blades, employs a combined multi-harmonic signal transmission signal, and utilizes time-frequency analysis methods to obtain the micro-Doppler features of blades with different numbers and rotational speeds. To more accurately identify and perceive different types of propellers, this invention constructs a CAMobileNetV2 model combining CBAM (Convolutional Block Attention Module) and MobileNetV2. This model can effectively analyze and extract the micro-Doppler features of the propeller, thereby achieving the identification of different underwater vehicles. Training and testing on the constructed dataset demonstrate the effectiveness and robustness of this technology in underwater sonar signal processing and underwater target recognition. This invention provides a micro-Doppler feature recognition technique in the field of underwater sonar signal processing, which is beneficial for improving the target recognition capability of underwater sonar.

Description

Technical Field

[0001] This invention belongs to the field of sonar signal processing, specifically involving an underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2. Technical Background

[0002] Underwater sonar target identification technology is an important research topic in the field of underwater acoustic signal processing. Traditional underwater target identification methods mostly rely on Doppler information carried by echo signals or underwater sonar imaging. During the movement of an underwater vehicle, the micro-Doppler effect caused by the minute frequency shifts resulting from the rotational motion of its propeller blades is also an important identification feature. Time-frequency analysis of the blade echo signals is performed, considering the number of propeller blades, their rotational speed, and the emission patterns of multiharmonic signals. The number of blades directly affects the spectral characteristics of the echo signal. Each blade generates a specific micro-Doppler frequency component during rotation; the more blades there are, the more frequency components in the echo signal. By analyzing these frequency components, information about the number of blades can be obtained. The rotational speed determines the frequency at which the blade passes through a specific spatial point, which is the frequency shift due to the micro-Doppler effect. High rotational speeds result in larger frequency shifts, while low rotational speeds result in smaller frequency shifts. By accurately measuring these frequency shifts, the propeller rotational speed can be inferred. By transmitting multi-harmonic signals, echo signals containing multiple frequency components can be obtained at the receiving end. The distribution and variation trends of different harmonic components in the time-frequency domain can reflect the motion state and geometric characteristics of the propeller blades. Complete micro-Doppler characteristics can be derived based on the number of propeller blades, rotational speed, and the combination of transmitted multi-harmonic signals.

[0003] In the field of deep learning, attention mechanisms have become one of the important techniques for improving model performance. By emphasizing important features and suppressing unimportant features, it enables models to better capture key information in the data. CBAM (Convolutional Block Attention Module) is an effective attention mechanism that combines channel attention and spatial attention, adaptively adjusting the importance of feature maps in both channel and spatial dimensions to further enhance model performance. MobileNetV2 is an efficient convolutional neural network architecture that employs depthwise separable convolutions and inverse residual structures to reduce model parameters and computational cost. This invention introduces the CBAM module based on MobileNetV2 to construct a CAMobileNetV2 fusion model, thereby improving the feature extraction capability and classification performance of the lightweight model. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and propose an underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2 to improve the underwater sonar target recognition capability in the prior art.

[0005] To achieve the above objectives, the present invention employs the following technical methods.

[0006] The underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2 includes the following steps:

[0007] Step 1: Use the multi-harmonic constant-frequency-modulated signal as the sonar's transmission signal and establish a transmission signal model;

[0008] Step 2: Model the underwater propeller blades and obtain the echo model and micro-Doppler information of the blades based on the sonar transmission signal.

[0009] Step 3: Perform time-frequency analysis on the echo signal to obtain the micro-Doppler time-frequency diagram;

[0010] Step 4: Use the time-frequency diagrams obtained by combining different numbers of blades, different blade rotation speeds, and different combinations of multiharmonic signals as the dataset.

[0011] Step 5: Divide the dataset into training and test sets proportionally;

[0012] Step 6: Combine MobileNetV2 with the attention mechanism to construct the CAMobileNetV2 model to improve the performance of the lightweight neural network. Perform micro-Doppler feature extraction on the input time-frequency map and fuse spatial and channel features to obtain a fused feature map for classification and recognition.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] (1) The micro-Doppler generated by the rotation of the propeller blades of the underwater vehicle is used as a feature for identification, which solves the problem that excessive reverberation in complex underwater scenes will cover the target features and can effectively extract the micro-motion information of the target.

[0015] (2) The transmission signal adopts a combination of multiple harmonics, which can ensure that the combination of echoes of different frequencies can carry more echo information, enhance the anti-reverberation performance, and improve the feature capture capability.

[0016] (3) The CAMobileNetV2 model can adapt to different environmental conditions and target types, making it more flexible in dealing with various underwater environments and diverse targets. The model can also learn to extract high-resolution features and capture subtle details that traditional methods may ignore. The CBAM attention mechanism is introduced into the model to improve feature extraction capabilities and improve the performance of micro-Doppler recognition features. Attached Figure Description

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0018] Figure 1 This is a flowchart of underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2.

[0019] Figure 2 This is a diagram of a propeller blade micro-motion model.

[0020] Figure 3 It is a time-frequency diagram of the simulated transmitted signal.

[0021] Figure 4 This is a diagram of the CAMobileNetV2 model.

[0022] Figure 5(a) is a microDoppler feature map when the number of leaves is 3.

[0023] Figure 5(b) is a microDoppler feature map when the number of blades is 4.

[0024] Figure 5(c) shows the micro-Doppler characteristics when the emitted harmonics are the first harmonic (S1) and the second harmonic (S2).

[0025] Figure 5(d) shows the micro-Doppler characteristics when the emitted harmonics are the first harmonic (S1) and the third harmonic (S3).

[0026] Figure 5(e) shows the micro-Doppler characteristics when the emitted harmonics are the first harmonic (S1) and the third harmonic (S4).

[0027] Figure 5(f) is a micro-Doppler feature map at a rotation speed of 5 m / s.

[0028] Figure 5(g) is a micro-Doppler feature map at a rotation speed of 6 m / s.

[0029] Figure 5(h) is a micro-Doppler feature map at a rotation speed of 7 m / s.

[0030] Figure 6 It is the model classification accuracy curve. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0032] like Figure 1 As shown, the underwater sonar multi-harmonic micro-Doppler feature recognition technology based on CAMobileNetV2 includes the following steps:

[0033] Step 1: Use the multi-harmonic constant-frequency-modulated signal as the sonar's transmission signal and establish a transmission signal model;

[0034] Specifically, step 1 includes the following sub-steps:

[0035] Sub-step 1.1: The sonar system transmits pulses consisting of four harmonics, each containing a long constant frequency (CF) component and a short frequency modulation (FM) component, totaling eight subcarriers, namely CF1-CF4 and FM1-FM4.

[0036] The time-domain expression of the CF component is:

[0037]

[0038] In the formula u i (t) represents the complex envelope of the signal, f cf The carrier frequency is represented by T, the signal duration is represented by i, and the number of harmonics is represented by i.

[0039] The FM component is a hyperbolic frequency modulation signal. Let the bandwidth of the transmitted signal be B, and the upper and lower frequency limits be respectively... and f fm The center frequency of the transmitted signal is denoted as:

[0040]

[0041] The time-domain expression for the FM component is:

[0042]

[0043] In the formula, k is the frequency modulation parameter, that is:

[0044]

[0045] r(t) represents the standard rectangular impulse function, which has the following form:

[0046]

[0047] Therefore, the phase and instantaneous frequency of the signal can be expressed as:

[0048]

[0049]

[0050] The fourth harmonic CF-FM signal emitted by the bionic sonar is:

[0051]

[0052] Step 2: Model the underwater propeller blades and obtain the echo model and micro-Doppler information of the blades based on the sonar transmission signal.

[0053] Sub-step 2.1 involves mathematically modeling the propeller blades, with the phase function being:

[0054]

[0055] Therefore, the echo signal of the blade is...

[0056]

[0057]

[0058] The echo is integrated into the blade length L, and the blade echo is:

[0059]

[0060]

[0061] The micro-Doppler representation of the echo signal is as follows:

[0062]

[0063] Step 3: Perform time-frequency analysis on the echo signal to obtain the micro-Doppler time-frequency diagram;

[0064] Specifically, step 3 includes the following sub-steps:

[0065] Sub-step 3.1 involves performing time-frequency analysis on the echo signal of the propeller blades using short-time Fourier transform to obtain a time-frequency map containing micro-Doppler information. The resulting time-frequency map includes the motion characteristics of the propeller blades at different times and frequencies, which can not only capture the changes in the rotational speed and angular velocity of the propeller blades, but also serve as a basis for classifying and identifying different propeller blades.

[0066] Step 4: Use the time-frequency diagrams obtained by combining different numbers of blades, different blade rotation speeds, and different combinations of multiharmonic signals as the dataset.

[0067] Specifically, step 4 includes the following sub-steps:

[0068] Sub-step 4.1: By changing the number of blades, the rotation speed of the blades, and the combination of transmitted signals, time-frequency analysis is performed on each echo signal to obtain a micro-Doppler time-frequency diagram;

[0069] Sub-step 4.2 integrates micro-Doppler features under different variable conditions, setting the input size of the time-frequency plot to 224*224. To increase the diversity of the training data, random cropping, flipping, rotation, and scaling techniques are used to augment the data, forming an underwater propeller blade micro-Doppler dataset.

[0070] Step 5: Divide the dataset into training and test sets proportionally;

[0071] Specifically, step 5 includes the following sub-steps:

[0072] Sub-step 5.1: Randomly select each type of echo from the propeller blade micro-Doppler feature dataset and combine them into a training set and a test set in an 8:2 ratio.

[0073] Step 6: Combine MobileNetV2 with the attention mechanism to construct the CAMobileNetV2 model to improve the performance of the lightweight neural network. Perform micro-Doppler feature extraction on the input time-frequency map and fuse spatial and channel features to obtain a fused feature map for classification and recognition.

[0074] Specifically, step 6 includes the following sub-steps:

[0075] Sub-step 6.1: Construct the channel attention module, generating an attention weight for each channel. Given the input feature map. Where C is the number of channels, and H and W are the height and width of the feature map, respectively. The calculation process of the channel attention module is as follows:

[0076] Calculate global average pooling and global max pooling to obtain two feature vectors describing the global information of the channels:

[0077] F avg =AvgPool(F), F max =MaxPool(F) (15)

[0078] Among them, F avg ,

[0079] The two feature vectors are each passed through a shared multilayer perceptron (MLP) and then summed:

[0080] M c =σ(MLP(F) avg )+MLP(F max (16)

[0081] Where σ represents the sigmoid function, It is a channel attention map.

[0082] Weight the input feature map using channel attention maps:

[0083] F′=M c ·F (17)

[0084] Sub-step 6.2 involves constructing a spatial attention module to generate an attention weight for each spatial location. Given the input feature map... The calculation process of the spatial attention module is as follows:

[0085] Calculate the average and max pooling of the feature maps along the channel dimension to obtain two feature maps describing spatial information:

[0086] F′ avg =AvgPool c (F′), F′ max =MaxPool c (F′) (18)

[0087] Among them, F′ avg ,

[0088] The two feature maps are concatenated along the channel dimension and passed through a convolutional layer:

[0089] M s =σ(Conv([F′) avg ;F′ max ])) (19)

[0090] Where σ represents the sigmoid function. It is a spatial attention map.

[0091] The input feature map is weighted using a spatial attention map:

[0092] F″=M s ·F′ (20)

[0093] At this point, the CBAM module is complete.

[0094] Through the two steps described above, the CBAM module can adaptively adjust the weights of the feature maps in both channel and spatial dimensions, thereby improving the model's representational capabilities.

[0095] Sub-step 6.3: Following the steps above, the channel attention mechanism module and the spatial attention mechanism module are fused to obtain the CBAM module. MobileNetV2 is used as the base network, which contains multiple inverse residual modules. A CBAM module is added after each inverse residual module to enhance its feature extraction capability, thus constructing a CAMobileNetV2 fusion model.

[0096] Sub-step 6.4: Shuffle the time-frequency feature maps in the partitioned training set to obtain the disordered training set;

[0097] In sub-step 6.5, the initial learning rate of the model is set to 0.001. The Adam optimizer is used to iteratively update the parameters of the fusion network, and the cross-entropy loss function is used, which is suitable for multi-class classification tasks. The training batch size is 32, and the number of training epochs is 50. A learning rate scheduler is used to gradually decrease the learning rate during training, and the learning rate is halved when the validation loss no longer decreases. He initialization is used for weight initialization, which is suitable for the ReLU activation function. To prevent overfitting, L2 regularization is used in the convolutional and fully connected layers. Early stopping is used to monitor the validation loss, and training is stopped early when the validation loss no longer decreases after 10 consecutive epochs.

[0098] Sub-step 6.6 involves training the model based on preset model parameters. The test set data is then input into the trained model, and its output is the classification results for different propeller blade types.

[0099] Table 1 Transmitted Signal Parameter Settings

[0100]

[0101] The multi-harmonic transmitted signal was simulated according to the parameters in Table 1, and the time-frequency diagram is shown below. Figure 3 As shown in Figure 5. When the number of propeller blades is 3, the micro-Doppler time-frequency diagram of the blade echo is shown in Figure 5(a). The micro-Doppler time-frequency diagram of the number of blades is shown in Figure 5(b). The number of blades can be distinguished from the micro-Doppler distribution in the time-frequency diagram. The echo Doppler of different harmonic combinations is shown in Figures 5(c), 5(d), and 5(e). By utilizing the characteristic that combinations of different frequency harmonics can carry echo information of different frequencies, the micro-Doppler information of the blade echo can be extracted more effectively. The micro-Doppler time-frequency diagrams generated by changes in blade rotation speed are shown in Figures 5(f), 5(g), and 5(h). The magnitude of the echo micro-Doppler caused by changes in rotation speed is also reflected in the time-frequency diagram. Therefore, the above differences can be used for neural network recognition.

[0102] The training process of the CAMobileNetV2 model and the size of each layer are as follows: Figure 3As shown. The input feature map size is 224*224. The model uses MobileNetV2 as the base network, which contains multiple inverse residual modules. A CBAM module is added after each inverse residual module to enhance feature extraction capabilities.

[0103] The implementation of the model includes training and testing phases.

[0104] Table 2 Dataset Parameter Settings

[0105]

[0106] The model's recognition performance was verified through computer simulation. Simulation data parameters are shown in Table 2, with 500 training iterations. Based on the dataset settings in Table 2, the datasets were divided into two categories: 3-bladed propellers and 4-bladed propellers. The rotational speed in each dataset was divided into groups of 0.01 m / s. Combining three multi-harmonic transmission signal formats, a total of 603 feature maps were generated for each dataset. These feature maps were then divided into training and testing sets in an 8:2 ratio and fed into the CAMobileNetV2 model for recognition. The recognition accuracy curve is shown below. Figure 6 As shown. To verify model stability and prevent overfitting, 10 sets of randomized trials were set up for identification. Figure 6 It can be seen that the proposed model has significantly improved recognition accuracy and stability compared with the traditional model, with an overall recognition accuracy of 97.45%, further demonstrating that the model has a high recognition rate for multi-harmonic micro-Doppler features.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for underwater sonar multi-harmonic micro-Doppler feature recognition based on CAMobileNetV2, the method comprising the following steps: Step 1: Use the multi-harmonic constant-frequency-modulated signal as the sonar's transmission signal and establish a transmission signal model; Step 2: Model the underwater propeller blades and obtain the echo model and micro-Doppler information of the blades based on the sonar transmission signal. Step 3: Perform time-frequency analysis on the echo signal to obtain the micro-Doppler time-frequency diagram; Step 4: Use the time-frequency diagrams obtained by combining different numbers of blades, different blade rotation speeds, and different combinations of multiharmonic signals as the dataset. Step 5: Divide the dataset into training and test sets proportionally; Step 6: Introduce the CBAM module, which consists of channel attention mechanism and spatial attention mechanism, into the MobileNetV2 network to construct the CAMobileNetV2 model. Use the CAMobileNetV2 model to extract micro-Doppler features from the input time-frequency map and fuse the spatial features and channel features to obtain a fused feature map for classification and recognition.

2. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 1, characterized in that... Combining micro-Doppler features from time-frequency analysis, target classification and recognition are achieved using the CAMobileNetV2 model. Step 1 includes the following sub-steps: Sub-step 1.1: The sonar system transmits pulses consisting of 4 harmonics. Each harmonic contains a long constant frequency (CF) component and a short frequency modulation (FM) component, which are divided into a total of 8 subcarriers, namely CF1-CF4 and FM1-FM4. The time-domain expression for the CF component is: where u i (t) represents the signal complex envelope, f cf represents the carrier frequency, T represents the signal duration, and i represents the number of harmonics; The FM component is a hyperbolic frequency modulation signal. Let the bandwidth of the transmitted signal be B, and the upper and lower frequency limits be respectively... and f fm The center frequency of the transmitted signal is denoted as: The time-domain expression for the FM component is: In the formula, k is the frequency modulation parameter, that is: r(t) represents the standard rectangular impulse function, which has the following form: Therefore, the phase and instantaneous frequency of the signal can be expressed as: The fourth harmonic CF-FM signal emitted by the bionic sonar is:

3. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 2, step 2 includes the following sub-steps: Sub-step 2.1: Mathematically model the propeller blades, and their phase function is: Therefore, the echo signal of the blade is: The echo is integrated into the blade length L, and the blade echo is: We can know that micro-Doppler is:

4. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 3, step 3 includes the following sub-steps: Sub-step 3.1: Short-time Fourier transform is used to perform time-frequency analysis on the echo signal of the propeller blade to obtain a time-frequency map containing micro-Doppler information. The obtained time-frequency map includes the motion characteristics of the propeller blade at different times and frequencies, which can capture the changes in the rotational speed and angular velocity of the propeller blade, and serve as the basis for classifying and identifying different propeller blades.

5. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 4, step 4 includes the following sub-steps: Sub-step 4.1: By changing the number of blades, the rotation speed of the blades, and the combination of transmitted signals, time-frequency analysis is performed on each echo signal to obtain a micro-Doppler time-frequency diagram; Sub-step 4.2 integrates micro-Doppler features under different variable conditions, sets the input size of the time-frequency map to 224*224, and uses random cropping, flipping, rotation and scaling methods to enhance the data in order to increase the diversity of training data, thus forming an underwater propeller blade micro-Doppler dataset.

6. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 5, step 5 includes the following sub-steps: Sub-step 5.1: Randomly select each type of echo from the propeller blade micro-Doppler feature dataset and combine them into a training set and a test set in an 8:2 ratio.

7. The underwater sonar multi-harmonic micro-Doppler feature recognition method based on CAMobileNetV2 according to claim 6, step 6 includes the following sub-steps: Sub-step 6.1: Construct a channel attention module, generating an attention weight for each channel, given the input feature map. Where C is the number of channels, and H and W are the height and width of the feature map, respectively. The calculation process of the channel attention module is as follows: Calculate global average pooling and global max pooling to obtain two feature vectors describing the global information of the channels: F avg = AvgPool(F), F max = MaxPool(F) in, The two feature vectors are each passed through a shared multilayer perceptron (MLP) and then summed: M c = σ(MLP(F avg )+ MLP(F max )) Where σ represents the sigmoid function, It is a channel attention map; Weight the input feature map using channel attention maps: F′=M c ·F Sub-step 6.2: Construct a spatial attention module, generating an attention weight for each spatial location, given the input feature map. The calculation process of the spatial attention module is as follows: Calculate the average and max pooling of the feature maps along the channel dimension to obtain two feature maps describing spatial information: F′ avg =AvgPool c (F′),F′ max =MaxPool c (F′) in, The two feature maps are concatenated along the channel dimension and passed through a convolutional layer: M s =σ(Conv([F′ avg ;F′ max ])) Where σ represents the sigmoid function. It is a spatial attention map; The input feature map is weighted using a spatial attention map: F″=M s ·F′ At this point, the CBAM module is complete; Through the above two steps, the CBAM module can adaptively adjust the weights of the feature maps in both channel and spatial dimensions, thereby improving the model's representational power. Sub-step 6.3: Based on the above steps, the channel attention mechanism module and the spatial attention mechanism module are fused to obtain the CBAM module. MobileNetV2 is used as the base network, which contains multiple inverse residual modules. The CBAM module is added after each inverse residual module to enhance its feature extraction capability and construct the CAMobileNetV2 fusion model. Sub-step 6.4: Shuffle the time-frequency feature maps in the partitioned training set to obtain the disordered training set; In sub-step 6.5, the initial learning rate of the model is set to 0.

001. The Adam optimizer is selected to iteratively update the parameters of the fusion network. The cross-entropy loss function is used, which is suitable for multi-class classification tasks. The training batch size is 32, and the number of training epochs is 50. A learning rate scheduler is used to gradually reduce the learning rate during training. When the validation loss no longer decreases, the learning rate is reduced by half. The weights are initialized using He initialization, which is suitable for the ReLU activation function. To prevent overfitting, L2 regularization is used in the convolutional and fully connected layers. Early stopping is used to monitor the validation loss. When the validation loss no longer decreases for 10 consecutive epochs, training is stopped early. Sub-step 6.6: Train the model according to the preset model parameters, input the test set data into the trained model, and the output is the classification results of different propeller blade types.

Images