Multi-band sonar cooperative complex environment target accurate identification system

Abstract

The invention, which relates to the technical field of sonar detection, discloses a multi-band sonar-coordinated complex environment target accurate identification system comprising a control center, a sonar front-end processing module, a multi-mode fusion module and a target identification module. The method is used for solving the problems of insufficient echo information utilization, poor sonar anti-interference capability and incomplete information identification in the prior art, and comprises the following steps: acquiring sonar images and sonar echoes in a target environment through a sonar front-end processing module, capturing interference source information in the target environment, and carrying out suppression processing; a sonar feature extraction unit extracts respective modal information features from sonar images and sonar echoes, a multi-modal feature fusion module performs feature fusion on the respective modal information features of the sonar images and the sonar echoes so as to generate sonar comprehensive features, and finally, a target recognition module performs target recognition according to the sonar comprehensive features. And the related information of the target object in the target environment is identified, so that accurate identification of the target in the complex environment is realized.

Description

Technical Field

[0001] The present invention relates to the field of sonar detection technology, and in particular to a system for accurately identifying targets in complex environments using multi-band sonar collaboration. Background Art

[0002] Sonar is currently the most effective device for detecting underwater targets. Image sonar is a sonar that can directly image acoustic detection information. It plays an indispensable role in many fields such as ocean development, marine scientific research, and underwater security. In particular, the widespread application of unmanned equipment in the marine field has put forward higher requirements for image sonar's autonomous identification of underwater scenes and targets, autonomous tracking, and anti-interference. Intelligent image sonar that can autonomously identify and track targets has received increasing attention. Traditional image sonar mainly relies on sonar imaging features for target recognition, but has the following problems:

[0003] 1. Insufficient use of echo information: Existing technologies fail to fully extract the target echo features from the sonar echo of the target object. Instead, they rely solely on the image features extracted from the image obtained by the image sonar for target recognition, resulting in a lack of target object recognition accuracy.

[0004] 2. Poor anti-interference ability: Noise and interference signals in complex underwater environments affect the quality of sonar echoes corresponding to target objects, resulting in subsequent target recognition failures;

[0005] 3. Multimodal data fusion: There is a lack of effective means to fuse target features of different modalities in sonar imaging features. When fusing target features of different modalities, the information expressed is often not comprehensive.

[0006] Therefore, there is an urgent need for a complex environment target recognition system that integrates multi-band sonar collaboration, multi-directional echo signal anti-interference and multi-modal feature fusion. Summary of the Invention

[0007] In order to solve the above problems, the purpose of the present invention is to provide a system for accurately identifying targets in complex environments using multi-band sonar collaboration.

[0008] The object of the present invention can be achieved by the following technical solutions: a multi-band sonar-coordinated complex environment target accurate identification system, comprising a control center, wherein the control center is communicatively connected to a sonar front-end processing module, a multimodal fusion module, and a target identification module;

[0009] The sonar front-end processing module consists of an intelligent sonar acquisition terminal, a sonar interference suppression unit, and a sonar feature extraction unit;

[0010] The intelligent sonar acquisition terminal is used to collect sonar images and sonar echoes in the target environment;

[0011] The sonar interference suppression unit is used to capture interference source information in the target environment and suppress the interference source information;

[0012] The sonar feature extraction unit is used to extract the corresponding modal information features from the sonar image and the sonar echo respectively;

[0013] The multimodal feature fusion module is used to fuse the modal information features of the sonar image and the sonar echo to generate a comprehensive sonar feature;

[0014] The target recognition module is used to identify relevant information of the target object in the target environment based on the sonar comprehensive characteristics.

[0015] Furthermore, the process of the intelligent sonar acquisition terminal collecting sonar images and sonar echoes in the target environment includes:

[0016] The intelligent sonar acquisition terminal is composed of a first acquisition node and a second acquisition node;

[0017] The first acquisition node is used to perform a sonar imaging operation on the target environment, thereby acquiring a sonar image in the target environment, and determining whether the imaging index of the sonar image meets expectations. If so, the current sonar imaging operation is completed; if not, the sonar imaging operation is repeated;

[0018] The second acquisition node is used to perform multi-directional echo extraction on the target environment, and then collect sonar echoes in the target environment.

[0019] Furthermore, the multi-directional echo extraction process includes:

[0020] The second acquisition node is composed of a plurality of sensor channels;

[0021] Each sensor channel performs beam transmission and echo reception, generates a sonar signal beam for each sensor channel through beam transmission, sets the beam transmission coverage angle and azimuth range for each sensor channel, and the transmission frequency corresponding to the sonar signal beam, and transmits the sonar signal beam of each sensor channel into the target environment;

[0022] When the sonar signal beam enters the target environment and collides with an obstacle, it is reflected to generate an echo signal, and the beam reflection coverage angle azimuth corresponding to the echo signal is recorded;

[0023] Each sensor channel is used to receive all echo signals whose beam reflection coverage angle azimuth is within the beam transmission coverage angle azimuth range, extract the sonar echo as the channel azimuth corresponding to each sensor channel, filter out all echo signals whose beam reflection coverage angle azimuth is not within the beam transmission coverage angle azimuth range, and then obtain the sonar echo of the channel azimuth corresponding to each sensor channel, and construct the sonar echo of multi-channel azimuth.

[0024] Furthermore, the sonar interference suppression unit captures interference source information in the target environment and suppresses the interference source information in the following process:

[0025] Constructing useful information frequency bands for performing sonar imaging operations and multi-directional echo extraction in the target environment respectively, using a sonar interference suppression unit to capture all information frequency bands in the target environment, and calculating the frequency band correlation coefficient between each information frequency band and the useful information frequency band for sonar imaging operations or the useful information frequency band for multi-directional echo extraction;

[0026] The frequency band correlation coefficient is denoted as τ;

[0027] Set the interference judgment threshold and record it as Gr;

[0028] When τ < Gr, it means that the information frequency band is weakly correlated with the useful information frequency band, and the current information frequency band is marked as an interference segment. When τ ≥ Gr, it means that the information frequency band is strongly correlated with the useful information frequency band, and the current information frequency band is marked as a similar frequency band to the useful information frequency band.

[0029] All interference fragments are summarized as the interference source information in the target environment. All similar frequency bands of the useful information frequency band are not processed. All interference information sources in the target environment are intercepted and suppressed.

[0030] Furthermore, the suppression process includes:

[0031] Unsupervised learning of interference source information that interferes with sonar images is performed. Noisy sonar images are obtained as training data. The autoencoder network is trained using this training data. The sonar images are segmented into several image frames. Starting from the starting image frame, each image frame is sequentially judged to determine whether there is interference source information.

[0032] If so, reconstruct the missing intermediate frame data caused by the interference through the adjacent image frames, use the mean square error between the reconstructed intermediate frame data and the original image frame corresponding data as the cost function training model, supplement the missing data corresponding to each image frame through the cost function training model, and after the missing data is supplemented, remove the interference source information corresponding to the image frame that caused the interference;

[0033] If not, no action is taken;

[0034] Perform frequency band filtering on the interference source information that interferes with sonar echoes.

[0035] Furthermore, the process of extracting the corresponding modal information features from the sonar image and the sonar echo by the sonar feature extraction unit includes:

[0036] Acquire historical sonar image data and historical sonar echo data, and construct historical image datasets and historical sonar datasets respectively; construct a first modality recognition model for identifying features corresponding to sonar images using a convolutional neural network; construct a second modality recognition model for identifying features corresponding to sonar echoes using unsupervised learning techniques; and train the first modality recognition model or the second modality recognition model based on the historical image dataset and the historical sonar dataset respectively;

[0037] Setting a model recognition rate servo range for each of the first modal recognition model and the second modal recognition model, and calculating the model recognition rates of each of the first modal recognition model and the second modal recognition model after each model training. If the respective model recognition rates are within the model recognition rate servo range, the first modal recognition model or the second modal recognition model is constructed;

[0038] Analyzing the collected sonar image in the target environment through the first modal recognition model, and extracting the corresponding modal information features as the shape information of the target object;

[0039] The sonar echoes collected in the target environment are analyzed by the second modal recognition model, and the corresponding modal information features extracted are the material information of the target object.

[0040] Furthermore, the multimodal feature fusion module fuses the modal information features of the sonar image and the sonar echo to generate the sonar comprehensive features. The process includes:

[0041] The process steps for performing the feature fusion include data preprocessing, single-modal feature extraction, multi-modal feature decision fusion and feature optimization;

[0042] Data preprocessing includes data alignment and feature normalization. Through data preprocessing, the modal information features of sonar images and sonar echoes are converted into their corresponding paradigm modal information features.

[0043] Performing single-modal feature extraction on the paradigm modal information features of the sonar image and the sonar echo, thereby obtaining the shape feature parameters of the paradigm modal information features corresponding to the sonar image, and obtaining the material feature parameters of the paradigm modal information features corresponding to the sonar echo;

[0044] By performing multimodal feature decision fusion on the shape feature parameters and material feature parameters, preliminary sonar comprehensive features are generated, and the preliminary sonar comprehensive features are optimized to obtain the final sonar comprehensive features.

[0045] Furthermore, the target recognition module identifies relevant information of the target object in the target environment based on the sonar comprehensive characteristics, including:

[0046] The final sonar comprehensive features are input into the target recognition module to identify all target objects in the target environment and obtain relevant information of the target objects, including the coordinate position of the target objects, the shape outline of the target objects, and the material composition of the target objects.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows: the sonar image and sonar echo in the target environment are collected by the intelligent sonar acquisition end, and the interference source information in the target environment is captured by the sonar interference suppression unit, and the interference source information is suppressed, thereby effectively preventing the noise and interference signals in the complex underwater environment from affecting the sonar echo quality of the target object; and the intelligent sonar acquisition end integrates the sonar image and sonar echo data, adopts multi-channel azimuth collection for the sonar echo, fully extracts the relevant features in the sonar echo, and lays a solid foundation for subsequent target recognition; the modal information features of the sonar image and the sonar echo are respectively fused through the multimodal feature fusion module, and then the sonar comprehensive features are generated to identify the relevant information of the target object in the target environment. By fusing the target features of different modalities, the expressed information is more comprehensive, thereby improving the accuracy of target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the present invention. DETAILED DESCRIPTION

[0049] like Figure 1 As shown, a complex environment target precision recognition system coordinated by multi-band sonar includes a control center, which is communicatively connected to a sonar front-end processing module, a multimodal fusion module, and a target recognition module;

[0050] The sonar front-end processing module consists of an intelligent sonar acquisition terminal, a sonar interference suppression unit, and a sonar feature extraction unit;

[0051] The intelligent sonar acquisition terminal is used to collect sonar images and sonar echoes in the target environment;

[0052] The sonar interference suppression unit is used to capture interference source information in the target environment and suppress the interference source information;

[0053] The sonar feature extraction unit is used to extract the corresponding modal information features from the sonar image and the sonar echo respectively;

[0054] The multimodal feature fusion module is used to fuse the modal information features of the sonar image and the sonar echo to generate a comprehensive sonar feature;

[0055] The target recognition module is used to identify relevant information of the target object in the target environment based on the sonar comprehensive characteristics.

[0056] It should be further explained that, in the specific implementation process, the process of the intelligent sonar acquisition terminal collecting sonar images and sonar echoes in the target environment includes:

[0057] The intelligent sonar acquisition terminal is composed of a first acquisition node and a second acquisition node;

[0058] The first acquisition node is used to perform a sonar imaging operation on the target environment, thereby acquiring a sonar image in the target environment, and determining whether the imaging index of the sonar image meets expectations. If so, the current sonar imaging operation is completed; if not, the sonar imaging operation is repeated;

[0059] The imaging indicators corresponding to sonar images include range resolution and azimuth resolution;

[0060] Set the corresponding limit intervals for distance resolution and azimuth resolution;

[0061] When the range resolution and azimuth resolution are both within their respective defined ranges, the imaging indicators of the sonar image meet expectations; otherwise, the imaging indicators of the sonar image do not meet expectations;

[0062] The second acquisition node is used to perform multi-directional echo extraction on the target environment, thereby collecting sonar echoes in the target environment; wherein the content of the multi-directional echo extraction is as follows:

[0063] The second acquisition node is composed of a plurality of sensor channels, and the plurality of sensor channels are numbered and denoted as i, i=1, 2, 3, ..., n, where n is a natural number greater than 0;

[0064] Each sensor channel performs beam transmission and echo reception;

[0065] Generate a sonar signal beam corresponding to each sensor channel through beam transmission, set the beam transmission coverage angle and azimuth range of each sensor channel, and the transmission frequency corresponding to the sonar signal beam, and transmit the sonar signal beam of each sensor channel into the target environment;

[0066] When the sonar signal beam enters the target environment and collides with an obstacle in the target environment, the obstacle reflects and generates an echo signal corresponding to the sonar signal beam, and records the beam reflection coverage angle and azimuth corresponding to the echo signal;

[0067] Each sensor channel is used to receive all echo signals whose beam reflection coverage angle azimuth is within the range of the beam transmission coverage angle azimuth, extract the sonar echo corresponding to the specific channel azimuth of each sensor channel, filter out all echo signals whose beam reflection coverage angle azimuth is not within the range of the beam transmission coverage angle azimuth, and then obtain the sonar echo corresponding to the channel azimuth of each sensor channel, and construct the sonar echo of multiple channel azimuths.

[0068] It should be further explained that, in a specific implementation process, the sonar interference suppression unit captures interference source information in the target environment and suppresses the interference source information in the following process:

[0069] The sonar interference suppression unit constructs useful information frequency bands corresponding to the sonar imaging operation and the multi-directional echo extraction in the target environment, sets an information acquisition period, and captures all information frequency bands in the target environment during the information acquisition period. The sonar interference suppression unit calculates a frequency band correlation coefficient between each information frequency band and the useful information frequency band of the sonar imaging operation or the useful information frequency band of the multi-directional echo extraction;

[0070] The frequency band correlation coefficient is recorded as τ, and the calculation formula of τ is as follows:

[0071] τ=Lc / L;

[0072] Wherein, Lc represents the information field length at which the similarity between a certain information frequency band in the target environment and the useful information frequency band corresponding to sonar imaging operation or multi-directional echo extraction is greater than 90%, and L represents the information field length of the useful information frequency band corresponding to sonar imaging operation or multi-directional echo extraction;

[0073] Set the interference judgment threshold and record it as Gr;

[0074] When τ < Gr, it means that the information frequency band is weakly correlated with the useful information frequency band, and the current information frequency band is marked as an interference segment. When τ ≥ Gr, it means that the information frequency band is strongly correlated with the useful information frequency band, and the current information frequency band is marked as a similar frequency band to the useful information frequency band.

[0075] All interference fragments are summarized as interference source information in the target environment, and all similar frequency bands of the useful information frequency band are not processed;

[0076] After the information acquisition period ends, an interference suppression period is set. During the interference suppression period, all interference information sources in the target environment are intercepted and suppressed. The suppression process is as follows:

[0077] Unsupervised learning of interference source information that interferes with sonar images is performed. Noisy sonar images are obtained as training data. The autoencoder network is trained using this training data. The sonar images are segmented into several image frames. Starting from the starting image frame, each image frame is sequentially judged to determine whether there is interference source information.

[0078] If so, reconstruct the missing intermediate frame data caused by the interference through the adjacent image frames, use the mean square error between the reconstructed intermediate frame data and the original image frame corresponding data as the cost function training model, supplement the missing data corresponding to each image frame through the cost function training model, and after the missing data is supplemented, remove the interference source information corresponding to the image frame that caused the interference;

[0079] If not, no action is taken;

[0080] The interference source information of the interference sonar echo is frequency-band filtered, and the first frequency band and the second frequency band are set. If the information frequency band of the current interference source information is in the first frequency band, the interference source information of the corresponding part of the information frequency band is retained as a comparison data set, and a comparison database is constructed based on the comparison data set. If the information frequency band of the current interference source information is in the second frequency band, the interference source information of the corresponding part of the information frequency band is directly filtered.

[0081] It should be noted that, compared with the second frequency band, the information frequency band corresponding to the first frequency band deviates more from the interference source information, that is, the information frequency band in the first frequency band is weakly correlated with the interference source information, and the information frequency band in the second frequency band is relatively strongly correlated with the interference source information. The information frequency band divided into the first frequency band is within a safe range, and the corresponding comparison data set is summarized and generated, and a comparison database is constructed. The comparison database is used as a means of subsequent filtering of relatively safe information frequency bands, thereby reducing the workload of judging the interference source information.

[0082] It should be further explained that, in a specific implementation process, the process of the sonar feature extraction unit extracting the corresponding modal information features from the sonar image and the sonar echo respectively includes:

[0083] Acquire historical sonar image data and construct a historical image dataset. Use a convolutional neural network to build a first modality recognition model that identifies features corresponding to sonar images. Then, divide the historical image dataset into a training set, a validation set, and a test set, and then train the first modality recognition model.

[0084] Setting a model recognition rate servo range corresponding to the first modal recognition model;

[0085] Counting the model recognition rate of the first modal recognition model after each model training; if the model recognition rate is within the model recognition rate servo range, the first modal recognition model is constructed;

[0086] Similarly, historical sonar echo data is obtained and a historical sonar dataset is constructed. A second modality recognition model that identifies the corresponding features of sonar echoes is constructed through unsupervised learning technology, and the second modality recognition model is trained based on the historical sonar dataset.

[0087] Setting the model recognition rate servo range corresponding to the second modal recognition model;

[0088] The model recognition rate of the second modality recognition model after each model training is calculated. If the model recognition rate is within the model recognition rate servo range, the second modality recognition model is constructed.

[0089] The sonar images collected in the target environment are analyzed using the first modal recognition model, and the corresponding modal information features extracted are the shape information of the target object. The sonar echoes collected in the target environment are analyzed using the second modal recognition model, and the corresponding modal information features extracted are the material information of the target object.

[0090] It should be further explained that, in the specific implementation process, the multimodal feature fusion module fuses the modal information features of the sonar image and the sonar echo to generate the sonar comprehensive features. The process includes:

[0091] The process steps for performing the feature fusion include data preprocessing, single-modal feature extraction, multi-modal feature decision fusion and feature optimization;

[0092] The data preprocessing includes data alignment and feature normalization, and the modal information features of the sonar image and the sonar echo are converted into their corresponding paradigm modal information features through data preprocessing;

[0093] Among them, data alignment includes time alignment, spatial alignment and motion compensation alignment;

[0094] Match the timestamps of sonar images and echo information with each other through time alignment;

[0095] By spatially aligning the position coordinates of each pixel in the sonar image and the physical position coordinates corresponding to the echo signal, the coordinates are mapped to the same three-dimensional coordinate system.

[0096] For a target object that is in a moving state in the target environment, the spatial offset of the target object (such as a ship) is corrected based on its motion trajectory using its inertial navigation data to achieve motion-compensated alignment;

[0097] Normalize the modal information features of the sonar image, specifically normalizing the edges, contours and other collective features of the sonar image into a standardized vector and normalizing it to the range of 0-1;

[0098] Normalize the modal information characteristics of the sonar echo, specifically logarithmically compressing the physical characteristics of the echo information, such as scattering intensity and spectrum energy;

[0099] Performing single-modal feature extraction on the paradigm modal information features of the sonar image and the sonar echo, thereby obtaining the shape feature parameters of the paradigm modal information features corresponding to the sonar image, and obtaining the material feature parameters of the paradigm modal information features corresponding to the sonar echo;

[0100] By performing multimodal feature decision fusion on the shape feature parameters and material feature parameters, preliminary sonar comprehensive features are generated, and the preliminary sonar comprehensive features are optimized to obtain the final sonar comprehensive features.

[0101] The detailed description of multimodal feature decision fusion is as follows:

[0102] Constructing feature vectors corresponding to the shape feature parameters and the material feature parameters respectively, obtaining vector dimensions of the feature vectors of the shape feature parameters and the material feature parameters respectively, and constructing classifiers corresponding to the shape feature parameters and the material feature parameters respectively;

[0103] Among them, the shape feature parameters correspond to the shape mode classifier, and the material feature parameters correspond to the material mode classifier. The shape mode classifier outputs the category probability P-shape, and the material mode classifier outputs the category probability P-material, and the final decision is constructed;

[0104] The final decision is recorded as P-final, and the final decision is expressed as follows:

[0105] P-final=ω1×P-shape+ω2×P-material;

[0106] Among them, ω1 and ω2 are set according to the modal reliability of the shape modal classifier and the material modal classifier, ω1+ω2=1, ω1>0, ω2>0;

[0107] The feature optimization of the preliminary sonar comprehensive features is performed through a joint objective function, which is denoted as Γ. The joint objective function Γ is expressed as follows:

[0108] Γ=λ1*Γ(detect)+λ2*Γ(classify)+λ3*Γ(reconstruct);

[0109] Where Γ(detect) is the target detection loss function, Γ(classify) is the material classification loss function, Γ(reconstruct) is the self-supervised reconstruction loss function, λ1, λ2, and λ3 are the function weights corresponding to the target detection loss function, material classification loss function, and self-supervised reconstruction loss function, respectively. λ1+λ2+λ3=1, and λ1, λ2, and λ3 are all greater than 0.

[0110] It should be further explained that, in the specific implementation process, the target recognition module identifies the relevant information of the target object in the target environment based on the comprehensive sonar characteristics, including:

[0111] The final sonar comprehensive features are input into the target recognition module to identify all target objects in the target environment and obtain relevant information of the target objects, including the coordinate position of the target objects, the shape outline of the target objects, and the material composition of the target objects.

[0112] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A multi-band sonar coordinated target recognition system for complex environments, including a control center, characterized by: The control center is communicatively connected to a sonar front-end processing module, a multimodal fusion module, and a target recognition module; The sonar front-end processing module consists of an intelligent sonar acquisition terminal, a sonar interference suppression unit, and a sonar feature extraction unit; The intelligent sonar acquisition terminal is used to collect sonar images and sonar echoes in the target environment; The sonar interference suppression unit is used to capture interference source information in the target environment and suppress the interference source information; The sonar feature extraction unit is used to extract the corresponding modal information features from the sonar image and the sonar echo respectively; The multimodal feature fusion module is used to fuse the modal information features of the sonar image and the sonar echo to generate a comprehensive sonar feature; The target recognition module is used to identify relevant information of the target object in the target environment based on the sonar comprehensive characteristics.

2. The multi-band sonar coordinated complex environment target precision identification system according to claim 1 is characterized in that: The process of collecting sonar images and sonar echoes in the target environment by the intelligent sonar acquisition terminal includes: The intelligent sonar acquisition terminal is composed of a first acquisition node and a second acquisition node; The first acquisition node is used to perform a sonar imaging operation on the target environment, thereby acquiring a sonar image in the target environment, and determining whether the imaging index of the sonar image meets expectations. If so, the current sonar imaging operation is completed; if not, the sonar imaging operation is repeated; The second acquisition node is used to perform multi-directional echo extraction on the target environment, and then collect sonar echoes in the target environment.

3. The multi-band sonar coordinated complex environment target precision identification system according to claim 2 is characterized in that: The process of multi-azimuth echo extraction includes: The second acquisition node is composed of a plurality of sensor channels; Each sensor channel performs beam transmission and echo reception, generates a sonar signal beam for each sensor channel through beam transmission, sets the beam transmission coverage angle and azimuth range for each sensor channel, and the transmission frequency corresponding to the sonar signal beam, and transmits the sonar signal beam of each sensor channel into the target environment; When the sonar signal beam enters the target environment and collides with an obstacle, it is reflected to generate an echo signal, and the beam reflection coverage angle azimuth corresponding to the echo signal is recorded; Each sensor channel is used to receive all echo signals whose beam reflection coverage angle azimuth is within the beam transmission coverage angle azimuth range, extract the sonar echo as the channel azimuth corresponding to each sensor channel, filter out all echo signals whose beam reflection coverage angle azimuth is not within the beam transmission coverage angle azimuth range, and then obtain the sonar echo of the channel azimuth corresponding to each sensor channel, and construct the sonar echo of multi-channel azimuth.

4. The multi-band sonar coordinated complex environment target precision identification system according to claim 3 is characterized by: The sonar interference suppression unit captures interference source information in the target environment and suppresses the interference source information in the following process: Constructing useful information frequency bands for performing sonar imaging operations and multi-directional echo extraction in the target environment respectively, using a sonar interference suppression unit to capture all information frequency bands in the target environment, and calculating the frequency band correlation coefficient between each information frequency band and the useful information frequency band for sonar imaging operations or the useful information frequency band for multi-directional echo extraction; The frequency band correlation coefficient is denoted as τ; Set the interference judgment threshold and record it as Gr; When τ < Gr, it means that the information frequency band is weakly correlated with the useful information frequency band, and the current information frequency band is marked as an interference segment. When τ ≥ Gr, it means that the information frequency band is strongly correlated with the useful information frequency band, and the current information frequency band is marked as a similar frequency band to the useful information frequency band. All interference fragments are summarized as the interference source information in the target environment. All similar frequency bands of the useful information frequency band are not processed. All interference information sources in the target environment are intercepted and suppressed.

5. The multi-band sonar coordinated complex environment target precision identification system according to claim 4 is characterized in that: The suppression process includes: Unsupervised learning of interference source information that interferes with sonar images is performed. Noisy sonar images are obtained as training data. The autoencoder network is trained using this training data. The sonar images are segmented into several image frames. Starting from the starting image frame, each image frame is sequentially judged to determine whether there is interference source information. If so, reconstruct the missing intermediate frame data caused by the interference through the adjacent image frames, use the mean square error between the reconstructed intermediate frame data and the original image frame corresponding data as the cost function training model, supplement the missing data corresponding to each image frame through the cost function training model, and after the missing data is supplemented, remove the interference source information corresponding to the image frame that caused the interference; If not, no action is taken; Perform frequency band filtering on the interference source information that interferes with sonar echoes.

6. The multi-band sonar coordinated complex environment target precision identification system according to claim 5 is characterized in that: The process of extracting the corresponding modal information features from the sonar image and sonar echo by the sonar feature extraction unit includes: Acquire historical sonar image data and historical sonar echo data, and construct historical image datasets and historical sonar datasets respectively; construct a first modality recognition model for identifying features corresponding to sonar images using a convolutional neural network; construct a second modality recognition model for identifying features corresponding to sonar echoes using unsupervised learning techniques; and train the first modality recognition model or the second modality recognition model based on the historical image dataset and the historical sonar dataset respectively; Setting a model recognition rate servo range for each of the first modal recognition model and the second modal recognition model, and calculating the model recognition rates of each of the first modal recognition model and the second modal recognition model after each model training. If the respective model recognition rates are within the model recognition rate servo range, the first modal recognition model or the second modal recognition model is constructed; Analyzing the collected sonar image in the target environment through the first modal recognition model, and extracting the corresponding modal information features as the shape information of the target object; The sonar echoes collected in the target environment are analyzed by the second modal recognition model, and the corresponding modal information features extracted are the material information of the target object.

7. The multi-band sonar coordinated complex environment target precision identification system according to claim 6 is characterized in that: The multimodal feature fusion module fuses the modal information features of the sonar image and sonar echo to generate the sonar comprehensive features. The process includes: The process steps for performing the feature fusion include data preprocessing, single-modal feature extraction, multi-modal feature decision fusion and feature optimization; Data preprocessing includes data alignment and feature normalization. Through data preprocessing, the modal information features of sonar images and sonar echoes are converted into their corresponding paradigm modal information features. Performing single-modal feature extraction on the paradigm modal information features of the sonar image and the sonar echo, thereby obtaining the shape feature parameters of the paradigm modal information features corresponding to the sonar image, and obtaining the material feature parameters of the paradigm modal information features corresponding to the sonar echo; By performing multimodal feature decision fusion on the shape feature parameters and material feature parameters, preliminary sonar comprehensive features are generated, and the preliminary sonar comprehensive features are optimized to obtain the final sonar comprehensive features.

8. The multi-band sonar coordinated complex environment target precision identification system according to claim 7 is characterized in that: The target recognition module identifies the relevant information of the target object in the target environment based on the comprehensive sonar characteristics, including the following process: The final sonar comprehensive features are input into the target recognition module to identify all target objects in the target environment and obtain relevant information of the target objects, including the coordinate position of the target objects, the shape outline of the target objects, and the material composition of the target objects.

Images