Seabed sonar image classification method and device, electronic equipment and storage medium
By using the combination method of frequency channel attention network and feature classification network in the subsea sonar image classification system, the problem of low classification accuracy of subsea sonar images in the prior art is solved, and a higher classification accuracy is achieved.
Patent Information
- Application Number
- CN202510078763.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-17
- Publication Date
- 2025-05-06
AI Technical Summary
The existing subsea sonar image classification method is not high enough in terms of accuracy and is difficult to effectively apply in actual engineering.
The combination method of frequency channel attention network and feature classification network is adopted to extract and classify the subsea sonar images through the convolution module, multi-spectral channel attention module and pulse coding module.
Through the combination of channel dimension expansion and channel attention and pulse coding, the feature extraction of the seabed sonar image is enriched and the classification accuracy is significantly improved.
Smart Images
Figure CN119942211A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image classification, and in particular to a method, device, electronic equipment and storage medium for classifying submarine sonar images. Background Art
[0002] The spiking neural network is considered to be a new type of third-generation artificial neural network, and its working method is closer to the working mechanism of the biological brain. Unlike traditional artificial neural networks (ANN, Artificial Neural Network), the spiking neural network is built based on a real biological neuron model. Its increased time dimension is conducive to more accurate biological simulation, more robust information expression and more efficient network performance. Precisely because the spiking neural network has a natural advantage in implementation principle, through the continuous further design and optimization of researchers, it can lay a theoretical foundation for the new generation of artificial intelligence models. At present, researchers are conducting optimization research on the neuron model, encoding method and network structure design in the spiking neural network.
[0003] In the pulse neural network training algorithm, it can be roughly divided into two categories: unsupervised learning and supervised learning. Each learning and training algorithm has its representative algorithm. In unsupervised learning, the algorithm represented by Spike Timing Dependent Plasticity (STDP) is the current mainstream algorithm. In supervised learning, ANN conversion SNN and substitute gradient back propagation algorithm are the current mainstream algorithms. By converting SNN by ANN, the SNN model is realized at the cost of losing some accuracy. The back propagation of the substitute gradient uses the method of approximate derivatives to solve the problem of non-differentiable pulse sequences. A back propagation algorithm combining time domain and space domain (Spatio-Temporal BackPropagation, STBP) is established, and a threshold-related batch processing method based on STBP, STBP-tdBN, is proposed in subsequent research to realize the training of deep SNN networks.
[0004] Most of the existing submarine sonar image classification methods are based on deep learning technologies such as CNN. Compared with SNN, CNN has defects such as occupying more memory and high power consumption. It faces problems such as high power consumption and difficult deployment, and thus is difficult to play a role in actual engineering. SNN has the advantages of low energy consumption and good sparsity.
[0005] However, the network structure of the existing SNN model is relatively simple and its feature extraction capability is insufficient, resulting in low accuracy in the task of seabed sonar image classification. Summary of the invention
[0006] The present invention provides a method, device, electronic equipment and storage medium for classifying submarine sonar images, which are used to solve the technical problem that the classification accuracy of existing submarine sonar image classification tasks is low.
[0007] The present invention provides a method for classifying submarine sonar images, which is applied to a submarine sonar image classification system; the submarine sonar image classification system comprises a frequency channel attention network and a feature classification network; the frequency channel attention network comprises a convolution module, a multi-spectral channel attention module and a pulse coding module, and the method comprises:
[0008] Obtain original seabed sonar images;
[0009] Input the original seafloor sonar image into the convolution module and output an expanded image;
[0010] Input the extended image into the multi-spectral channel attention module and output a preprocessed feature map;
[0011] Input the preprocessing feature map into the pulse encoding module and output a pulse feature map;
[0012] The pulse feature map is input into a feature classification network to obtain feature classification of the seabed sonar image.
[0013] Optionally, the step of inputting the extended image into the multi-spectral channel attention module and outputting a preprocessed feature map comprises:
[0014] By means of the multi-spectral channel attention module, the extended image is segmented according to channels to obtain multiple frequency components;
[0015] Performing a two-dimensional discrete cosine transform on each of the frequency components to obtain a plurality of transform feature graphs;
[0016] Each of the transformed feature maps is multiplied by the expanded image according to the channel to obtain a preprocessed feature map.
[0017] Optionally, the step of inputting the preprocessing feature map into the pulse encoding module and outputting the pulse feature map comprises:
[0018] The preprocessing feature map is encoded by the pulse encoding module to generate a pulse feature map.
[0019] Optionally, the feature classification network includes a convolution layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; the step of inputting the pulse feature map into the feature classification network to obtain the feature classification of the seabed sonar image includes:
[0020] Inputting the pulse feature map into the convolution layer and the maximum pooling layer to obtain a first-size compressed image;
[0021] Inputting the first-size compressed image into the first multi-scale residual module to obtain a second-size compressed image;
[0022] Inputting the second-size compressed image into the second multi-scale residual module to obtain a third-size compressed image;
[0023] The third-size compressed image is input into the fully connected layer, and the feature classification of the seabed sonar image is output.
[0024] The present invention also provides a submarine sonar image classification device, which is applied to a submarine sonar image classification system; the submarine sonar image classification system includes a frequency channel attention network and a feature classification network; the frequency channel attention network includes a convolution module, a multi-spectral channel attention module and a pulse coding module, and the device includes:
[0025] The original seabed sonar image acquisition module is used to acquire the original seabed sonar image;
[0026] An extended image output module, used to input the original seabed sonar image into the convolution module and output an extended image;
[0027] A preprocessing feature map output module, used for inputting the extended image into the multi-spectral channel attention module and outputting a preprocessing feature map;
[0028] A pulse characteristic diagram output module, used for inputting the preprocessing characteristic diagram into the pulse encoding module and outputting a pulse characteristic diagram;
[0029] The feature classification module is used to input the pulse feature map into a feature classification network to obtain feature classification of the seabed sonar image.
[0030] Optionally, the preprocessing feature map output module includes:
[0031] A segmentation submodule, used to segment the extended image according to channels through the multi-spectral channel attention module to obtain multiple frequency components;
[0032] A transformation submodule, used for performing a two-dimensional discrete cosine transform on each of the frequency components to obtain a plurality of transformation feature maps;
[0033] The preprocessing feature map generation submodule is used to multiply each of the transformed feature maps and the extended image according to the channel to obtain the preprocessing feature map.
[0034] Optionally, the pulse characteristic diagram output module includes:
[0035] The pulse characteristic diagram output submodule is used to encode the preprocessing characteristic diagram through the pulse encoding module to generate a pulse characteristic diagram.
[0036] Optionally, the feature classification network includes a convolution layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; the feature classification module includes:
[0037] A first-size compressed image generation submodule, used for inputting the impulse feature map into the convolution layer and the maximum pooling layer to obtain a first-size compressed image;
[0038] A second-size compressed image generating submodule, configured to input the first-size compressed image into the first multi-scale residual module to obtain a second-size compressed image;
[0039] A third-size compressed image generating submodule, configured to input the second-size compressed image into the second multi-scale residual module to obtain a third-size compressed image;
[0040] The feature classification submodule is used to input the third-size compressed image into the fully connected layer and output the feature classification of the seabed sonar image.
[0041] The present invention also provides an electronic device, the device comprising a processor and a memory:
[0042] The memory is used to store program code and transmit the program code to the processor;
[0043] The processor is used to execute the seabed sonar image classification method as described in any one of the above items according to the instructions in the program code.
[0044] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the seabed sonar image classification method as described in any one of the above items.
[0045] It can be seen from the above technical solutions that the present invention has the following advantages: The present invention provides a method for classifying seabed sonar images, and specifically discloses: obtaining an original seabed sonar image; inputting the original seabed sonar image into a convolution module to output an expanded image; inputting the expanded image into a multispectral channel attention module to output a preprocessed feature map; inputting the preprocessed feature map into a pulse coding module to output a pulse feature map; inputting the pulse feature map into a feature classification network to obtain feature classification of the seabed sonar image. The present invention enriches the feature extraction of the original image by expanding the channel dimension of the seabed sonar image and then combining channel attention with pulse coding, thereby improving the classification accuracy of the seabed sonar image. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0047] Figure 1 A flowchart of a method for classifying submarine sonar images provided by an embodiment of the present invention;
[0048] Figure 2 A schematic diagram of the structure of a frequency channel attention network provided by an embodiment of the present invention;
[0049] Figure 3 A flowchart of a method for classifying seabed sonar images provided by another embodiment of the present invention;
[0050] Figure 4 A schematic diagram of image compression based on two-dimensional discrete cosine transform provided by an embodiment of the present invention;
[0051] Figure 5 A schematic diagram of the results of the feature classification network model provided by an embodiment of the present invention;
[0052] Figure 6 A flowchart of executing a multi-scale residual module provided in an embodiment of the present invention;
[0053] Figure 7 A structural block diagram of a submarine sonar image classification device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0054] The embodiments of the present invention provide a method, device, electronic device and storage medium for classifying submarine sonar images, which are used to solve the technical problem that the classification accuracy of existing submarine sonar image classification tasks is low.
[0055] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0056] See also Figure 1 , Figure 1 A flowchart of the steps of a method for classifying seabed sonar images provided by an embodiment of the present invention.
[0057] The present invention provides a method for classifying submarine sonar images, which is applied to a submarine sonar image classification system; the submarine sonar image classification system includes a frequency channel attention network and a feature classification network; the frequency channel attention network includes a convolution module, a multispectral channel attention module and a pulse coding module. Among them, the frequency channel attention network FcaNet can automatically learn the key features in the image, especially in the case of low contrast and strong noise interference, and shows high robustness.
[0058] The method may specifically include the following steps:
[0059] Step 101, obtaining an original seabed sonar image;
[0060] Sonar, also known as sonar, is a technology that uses the propagation and reflection characteristics of sound waves in water to navigate and measure distance through electro-acoustic conversion and information processing.
[0061] By analyzing the underwater sonar images, we can determine what is on the seabed, such as crashed planes, sunken ships, victims, and the seabed.
[0062] Step 102, inputting the original seafloor sonar image into a convolution module, and outputting an expanded image;
[0063] like Figure 2 As shown, in an embodiment of the present invention, the original seabed sonar image (such as 1×56×56 pixels) can be input into a convolution module (such as 1×1 convolution) to expand the original seabed sonar image to 16 channels, and the output is a 16×56×56 expanded image.
[0064] Step 103, input the expanded image into a multi-spectral channel attention module, and output a pre-processed feature map;
[0065] like Figure 2 As shown in FIG. 1 , after obtaining the extended image, the extended image can be input into the multi-spectral channel attention module to generate a preprocessed feature map of 16×56×56 pixels.
[0066] Step 104, inputting the preprocessing characteristic graph into a pulse coding module, and outputting a pulse characteristic graph;
[0067] The 16×56×56 pixel preprocessed feature map output by the multi-spectral channel attention module can be regarded as the membrane potential of the multi-feature input, which is encoded by the pulse encoding module and finally outputs a 16×56×56 pulse feature map.
[0068] Step 105, inputting the pulse feature map into a feature classification network to obtain feature classification of the seabed sonar image.
[0069] In the embodiment of the present invention, the core of the feature classification network is a multi-scale residual module. The pulse feature map is input into the feature classification network and can be classified to obtain corresponding feature classifications.
[0070] The present invention enriches the feature extraction of the original image by expanding the channel dimension of the seabed sonar image and then combining channel attention with pulse coding, thereby improving the classification accuracy of the seabed sonar image.
[0071] See also Figure 3 , Figure 3 A flowchart of a method for classifying seabed sonar images provided by another embodiment of the present invention. Specifically, the following steps may be included:
[0072] Step 301, obtaining an original seabed sonar image;
[0073] Step 302, inputting the original seafloor sonar image into a convolution module, and outputting an expanded image;
[0074] Steps 301-302 are the same as steps 101-102. For details, please refer to the description of steps 101-102, which will not be repeated here.
[0075] Step 303, segmenting the extended image according to channels through a multi-spectral channel attention module to obtain multiple frequency components;
[0076] Step 304, performing a two-dimensional discrete cosine transform on each frequency component to obtain a plurality of transform feature graphs;
[0077] Step 305, multiplying each transformed feature map with the expanded image according to the channel to obtain a preprocessed feature map;
[0078] In an embodiment of the present invention, the expanded image can be divided into multiple frequency components along the channel direction. Then, it is compressed by two-dimensional discrete cosine transform (e.g. Figure 4 As shown in Figure 2), multiple transformed feature maps are obtained. Finally, the corresponding channels are multiplied with the expanded image entering the multi-spectral channel attention module, and the obtained 16×56×56 feature map is the pre-processed feature map after the multi-spectral channel attention processing.
[0079] The two-dimensional discrete cosine transform (DCT) is a transform related to the Fourier transform. It is similar to the discrete Fourier transform (DFT), but only uses real numbers. The discrete cosine transform is equivalent to a discrete Fourier transform that is about twice as long. This discrete Fourier transform is performed on a real even function (because the Fourier transform of a real even function is still a real even function). In some transformations, the input or output position needs to be moved by half a unit.
[0080] The mathematical expression of two-dimensional discrete cosine transform (DCT) is as follows:
[0081] The basis functions of the two-dimensional DCT are:
[0082]
[0083] The above represents a basis function for a specific frequency, where i and j represent the row and column indices of the frequency, and h and w represent the spatial position. H and W represent the height and width of the feature map, respectively. For each spatial position (h, w) in the feature map, there is a basis function corresponding to the frequency component specified by i and j. The value of this basis function depends on the spatial position and the frequency component. The purpose of the basis function is to decompose the signal in the spatial domain into the sum of different frequency components.
[0084] The two-dimensional DCT is expressed as:
[0085]
[0086] The above formula describes the process of DCT transformation, which transforms the signal in the spatial domain Converted to the frequency domain, h and w represent the row index and column index of the frequency component, which are used to distinguish different frequency components. i and j represent the spatial position, which are used to calculate the value of a specific frequency component at each spatial position.
[0087] The inverse two-dimensional DCT is expressed as:
[0088]
[0089] Step 306, encoding the preprocessing feature map through a pulse encoding module to generate a pulse feature map;
[0090] In an embodiment of the present invention, the 16×56×56 pixel preprocessed feature map output by the multi-spectral channel attention module can be regarded as the membrane potential of the multi-feature input, which is encoded by the pulse coding module and finally outputs a 16×56×56 pulse feature map.
[0091] Step 307, input the pulse feature map into the feature classification network to obtain feature classification of the seabed sonar image.
[0092] In the embodiment of the present invention, the core of the feature classification network is a multi-scale residual module. The pulse feature map is input into the feature classification network and can be classified to obtain corresponding feature classifications.
[0093] In one example, the feature classification network includes a convolution layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; the step of inputting the pulse feature map into the feature classification network to obtain the feature classification of the seabed sonar image may specifically include the following sub-steps:
[0094] S71, inputting the pulse feature map into the convolution layer and the maximum pooling layer to obtain a first size compressed image;
[0095] S72, inputting the first-size compressed image into a first multi-scale residual module to obtain a second-size compressed image;
[0096] S73, inputting the second-size compressed image into a second multi-scale residual module to obtain a third-size compressed image;
[0097] S74, input the compressed image of the third size into the fully connected layer, and output the feature classification of the seabed sonar image.
[0098] In the specific implementation, Figure 5 As shown in the figure, the core of the feature classification network is the multi-scale residual module. In this network structure, the input image is a 16×56×56 pixel pulse feature map. After a 3×3 standard convolution layer and a maximum pooling layer, the output is reduced to half of the original input. Then it passes through two multi-scale residual modules (the first multi-scale residual module and the second multi-scale residual module). Each multi-scale residual module is connected to a maximum pooling layer to reduce the size of the feature map. After passing through all modules, the feature classification of the seabed sonar image is output after passing through the fully connected layer.
[0099] Among them, the execution process of the multi-scale residual module is as follows Figure 6As shown in the figure, by using three different sizes of convolution kernels to extract multi-scale features, the representation ability of the network model can be effectively improved. At the same time, connecting the three different sizes of convolution kernels in parallel can avoid overfitting caused by excessive network stacking. Adding a 1×1 convolution kernel before the convolution kernels of each size can reduce the dimension of the features. The 1×1 convolution kernel added after the filter connection increases the dimension of the merged features. By adding a 1×1 convolution kernel, the number of parameters in the network is reduced. In addition, a batch normalization layer (BN, Btach Normalization) is added to the pulse activation layer after each convolution layer. The role of this method is to make the input data distribution transformation more robust by normalizing each batch of data, so as to avoid the problem of increased network calculation and prolonged training time due to the difference in the distribution of input image sample data in each batch in back propagation, thereby accelerating the convergence of network training.
[0100] The present invention enriches the feature extraction of the original image by expanding the channel dimension of the submarine sonar image and then combining channel attention with pulse coding, thereby improving the classification accuracy of the submarine sonar image. In addition, by using a pulse multi-scale residual network composed of multiple convolutional layers and residual modules of different scales, the inaccuracy caused by a single convolution when extracting features is solved, and the receptive field of the network is also expanded.
[0101] See also Figure 7 , Figure 7 A structural block diagram of a submarine sonar image classification device provided in an embodiment of the present invention.
[0102] The embodiment of the present invention provides a submarine sonar image classification device, which is applied to a submarine sonar image classification system; the submarine sonar image classification system includes a frequency channel attention network and a feature classification network; the frequency channel attention network includes a convolution module, a multi-spectral channel attention module and a pulse coding module, and the device includes:
[0103] The original seabed sonar image acquisition module 701 is used to acquire the original seabed sonar image;
[0104] The expanded image output module 702 is used to input the original seabed sonar image into the convolution module and output the expanded image;
[0105] A preprocessing feature map output module 703 is used to input the extended image into the multi-spectral channel attention module and output the preprocessing feature map;
[0106] The pulse characteristic diagram output module 704 is used to input the preprocessing characteristic diagram into the pulse encoding module and output the pulse characteristic diagram;
[0107] The feature classification module 705 is used to input the pulse feature map into the feature classification network to obtain the feature classification of the seabed sonar image.
[0108] In the embodiment of the present invention, the preprocessing feature map output module 703 includes:
[0109] The segmentation submodule is used to segment the expanded image according to channels through the multi-spectral channel attention module to obtain multiple frequency components;
[0110] A transformation submodule, used for performing a two-dimensional discrete cosine transform on each frequency component to obtain a plurality of transformation feature maps;
[0111] The preprocessing feature map generation submodule is used to multiply each transformed feature map with the expanded image according to the channel to obtain the preprocessing feature map.
[0112] In the embodiment of the present invention, the pulse characteristic diagram output module 704 includes:
[0113] The pulse feature map output submodule is used to encode the preprocessing feature map through the pulse encoding module to generate a pulse feature map.
[0114] In the embodiment of the present invention, the feature classification network includes a convolution layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; the feature classification module 705 includes:
[0115] A first-size compressed image generation submodule is used to input the pulse feature map into the convolution layer and the maximum pooling layer to obtain a first-size compressed image;
[0116] A second size compressed image generation submodule, used for inputting the first size compressed image into the first multi-scale residual module to obtain a second size compressed image;
[0117] A third-size compressed image generation submodule, used for inputting the second-size compressed image into the second multi-scale residual module to obtain a third-size compressed image;
[0118] The feature classification submodule is used to input the third-size compressed image into the fully connected layer and output the feature classification of the seabed sonar image.
[0119] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:
[0120] The memory is used to store the program code and transmit the program code to the processor;
[0121] The processor is used to execute the seabed sonar image classification method of the embodiment of the present invention according to the instructions in the program code.
[0122] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the seabed sonar image classification method of the embodiment of the present invention.
[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0124] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0125] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0126] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0127] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0128] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0129] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.
[0130] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.
[0131] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. 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 method for classifying submarine sonar images, characterized in that: Applied to a submarine sonar image classification system; the submarine sonar image classification system comprises a frequency channel attention network and a feature classification network; The frequency channel attention network includes a convolution module, a multi-spectral channel attention module and a pulse coding module, and the method includes: Obtain original seabed sonar images; Input the original seafloor sonar image into the convolution module and output an expanded image; Input the extended image into the multi-spectral channel attention module and output a preprocessed feature map; Input the preprocessing feature map into the pulse encoding module and output a pulse feature map; The pulse feature map is input into a feature classification network to obtain feature classification of the seabed sonar image.
2. The method according to claim 1, characterized in that The step of inputting the extended image into the multi-spectral channel attention module and outputting the preprocessed feature map comprises: By means of the multi-spectral channel attention module, the extended image is segmented according to channels to obtain multiple frequency components; Performing a two-dimensional discrete cosine transform on each of the frequency components to obtain a plurality of transform feature graphs; Each of the transformed feature maps is multiplied by the expanded image according to the channel to obtain a preprocessed feature map.
3. The method according to claim 1, characterized in that The step of inputting the preprocessing characteristic graph into the pulse coding module and outputting the pulse characteristic graph comprises: The preprocessing feature map is encoded by the pulse encoding module to generate a pulse feature map.
4. The method according to claim 1, characterized in that The feature classification network includes a convolution layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; the step of inputting the pulse feature map into the feature classification network to obtain the feature classification of the seabed sonar image includes: Inputting the pulse feature map into the convolution layer and the maximum pooling layer to obtain a first-size compressed image; Inputting the first-size compressed image into the first multi-scale residual module to obtain a second-size compressed image; Inputting the second-size compressed image into the second multi-scale residual module to obtain a third-size compressed image; The third-size compressed image is input into the fully connected layer, and the feature classification of the seabed sonar image is output.
5. A submarine sonar image classification device, characterized in that: Applied to a submarine sonar image classification system; the submarine sonar image classification system comprises a frequency channel attention network and a feature classification network; The frequency channel attention network includes a convolution module, a multi-spectral channel attention module and a pulse coding module, and the device includes: The original seabed sonar image acquisition module is used to acquire the original seabed sonar image; An extended image output module, used to input the original seabed sonar image into the convolution module and output an extended image; A preprocessing feature map output module, used for inputting the extended image into the multi-spectral channel attention module and outputting a preprocessing feature map; A pulse characteristic diagram output module, used for inputting the preprocessing characteristic diagram into the pulse encoding module and outputting a pulse characteristic diagram; The feature classification module is used to input the pulse feature map into a feature classification network to obtain feature classification of the seabed sonar image.
6. The device according to claim 5, characterized in that The preprocessing feature map output module comprises: A segmentation submodule, configured to segment the extended image according to channels through the multi-spectral channel attention module to obtain multiple frequency components; A transformation submodule, used for performing a two-dimensional discrete cosine transform on each of the frequency components to obtain a plurality of transformation feature maps; The preprocessing feature map generation submodule is used to multiply each of the transformed feature maps with the expanded image according to the channel to obtain the preprocessing feature map.
7. The device according to claim 5, characterized in that The pulse characteristic diagram output module comprises: The pulse characteristic diagram output submodule is used to encode the preprocessing characteristic diagram through the pulse encoding module to generate a pulse characteristic diagram.
8. The device according to claim 5, characterized in that The feature classification network includes a convolutional layer and a maximum pooling layer, a first multi-scale residual module, a second multi-scale residual module and a fully connected layer; The feature classification module comprises: A first-size compressed image generation submodule, used for inputting the impulse feature map into the convolution layer and the maximum pooling layer to obtain a first-size compressed image; A second-size compressed image generating submodule, configured to input the first-size compressed image into the first multi-scale residual module to obtain a second-size compressed image; A third-size compressed image generating submodule, configured to input the second-size compressed image into the second multi-scale residual module to obtain a third-size compressed image; The feature classification submodule is used to input the third-size compressed image into the fully connected layer and output the feature classification of the seabed sonar image.
9. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the seabed sonar image classification method described in any one of claims 1-4 according to the instructions in the program code.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program code, and the program code is used to execute the seabed sonar image classification method described in any one of claims 1-4.