CN117434524A - Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image - Google Patents
Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image Download PDFInfo
- Publication number
- CN117434524A CN117434524A CN202311321576.3A CN202311321576A CN117434524A CN 117434524 A CN117434524 A CN 117434524A CN 202311321576 A CN202311321576 A CN 202311321576A CN 117434524 A CN117434524 A CN 117434524A
- Authority
- CN
- China
- Prior art keywords
- interest
- echo data
- synthetic aperture
- small
- small object
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000004458 analytical method Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 12
- 230000001131 transforming effect Effects 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 32
- 230000006835 compression Effects 0.000 claims description 13
- 238000007906 compression Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- 238000010183 spectrum analysis Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 description 10
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 5
- 230000009466 transformation Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 241001481833 Coryphaena hippurus Species 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000945 filler Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/52001—Auxiliary means for detecting or identifying sonar signals or the like, e.g. sonar jamming signals
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/539—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Multimedia (AREA)
- Biophysics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention provides a method for identifying the attribute of echo data of a small object of interest in a synthetic aperture sonar image. The method comprises the following steps: processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method; correlating the position information of the small object of interest with the echo data to obtain the echo data of the small object of interest; transforming echo data of a small object of interest into voiceprint features; inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of a sonar image; the classification recognition model employs a modified MobileNet V3 network. According to the method, attribute analysis is carried out on the underwater small object of interest in the echo domain, the interpretability of the underwater small object of interest is improved, and an effective solution means is provided for the underwater object fine recognition task based on the SAS image.
Description
Technical Field
The invention relates to the field of underwater acoustic signal processing, in particular to a method for identifying the attribute of echo data of a small target of interest of a synthetic aperture sonar image.
Background
The synthetic aperture sonar (Synthetic Aperture Sonar, SAS) is a high-resolution underwater imaging sonar, and the basic principle is that a virtual large aperture is formed by using the movement of a small aperture matrix, so that high resolution in the azimuth direction is obtained. Compared with the common side-scan sonar, the SAS has the most remarkable advantages that the azimuth resolution is higher, and the theoretical resolution is irrelevant to the target distance and the adopted sound wave frequency band. The synthetic aperture sonar image target detection task plays an important role in autonomous navigation and search of an underwater unmanned platform. The underwater small targets of the synthetic aperture sonar images have the problems of intra-class difference and inter-class similarity, and under the condition of lacking priori knowledge, the underwater small targets are difficult to confirm only from an image domain. The array element echo data contains feature information rich in targets, and further analysis of small targets of interest is expected to be realized by means of the echo data.
The underwater target echo signal has a complex sound scattering component time sequence structure, and various sound scattering components have different frequency characteristics, so that the underwater target echo signal is suitable for analyzing the signal characteristics of the target sound scattering by adopting a time-frequency analysis method. The Gaunaord uses Cohen type time-frequency analysis method to study the echo signal characteristics of the underwater elastic target, and analyzes the relation between the target attribute parameters such as size, shape, shell thickness, material, internal filler and the like and the time-frequency characteristics of the target echo signal. Muller et al continue to study the time-frequency characteristic differences of different material elastic target echo signals when dolphin sound is used as an active sonar emission signal. Although a certain research result is obtained in the time-frequency characteristic direction of the target sound scattering signal, a lot of defects still exist, and the main problem is that the aliasing problem of various sound scattering components in the target echo signal cannot be solved.
Compared with the traditional machine learning and signal processing method, the deep learning simulates a hierarchical system of a human visual nervous system, contains more hidden unit layers, can obtain higher-level and more abstract feature expression through nonlinear transformation of original data layer by layer, and can strengthen the distinguishing capability of input data and weaken the adverse effect of irrelevant factors. And the time-frequency characteristics of the echo data are analyzed and processed by using a deep learning technology, so that the identification of the attribute of the underwater small target is possible. Howard et al propose a lightweight convolutional neural network MobileNet V1.MobileNet V1 replaces the standard convolution with a depth separable convolution (depthwise separable convolution, DSC) to reduce the parameters and computational effort of the model. Sandler et al propose a modified version of MobileNet V2 of MobileNet V1. The MobileNet V2 introduces a cross-connection (shortcut connections) structure on the basis of a depth separable convolution and designs a new feature extraction module IRB (inverted residual block). The new module adjusts the original 'compression' and 'expansion' to 'expansion' and 'compression', and simultaneously changes the active layer of the last convolution layer from non-linearity to linearity in order to reduce the loss and damage of the active function when the high-dimensional information is converted into the low-dimensional information. Howard et al propose an improved version of MobileNet V3 and feature extraction Module IRB on the basis of MobileNet V1 and MobileNet V2 + ，IRB + A SE (squeeze and excitation) attention mechanism was introduced. The SE firstly performs squeeze operation on the features obtained by convolution to obtain global features, then performs extraction operation on the global features to obtain weights of different features, and finally multiplies the weights of the corresponding channels to obtain final features. Essentially, the SE component selects in the feature dimension, which can be more focused on the maximum amount of informationWhile suppressing those that are not. IRB (IRB) + The method has better feature extraction capability while keeping low calculation amount. However, IRB + Capturing the dependencies of all channels is inefficient and unnecessary.
In view of the foregoing, a method suitable for identifying the attribute of the echo data of the small object of interest under water is urgently needed at present, so as to improve the accuracy and efficiency of the fine identification of the object.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a method for identifying the echo data attribute of a small object of interest of a synthetic aperture sonar image, so that the small object of interest under water is further confirmed.
In order to achieve the above object, the present invention provides a method for identifying the attribute of echo data of a small object of interest in a synthetic aperture sonar image, which comprises:
processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
correlating the position information of the small object of interest with the echo data to obtain the echo data of the small object of interest;
transforming echo data of a small object of interest into voiceprint features;
inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of a sonar image;
the classification recognition model employs a modified MobileNet V3 network.
Preferably, the transforming the echo data of the small object of interest into voiceprint features; the method specifically comprises the following steps:
preprocessing the echo data of the small object of interest, including sampling and quantization;
performing fast Fourier transform or other spectrum analysis technology on the preprocessed echo data, and converting the echo data into a spectrum signal;
and converting the frequency spectrum signal into a two-dimensional image of time and frequency to obtain the voiceprint feature map.
Preferably, the improved MobileNet V3 network adopts an EIRB module to replace IRB in the original MobileNet V3 + A module; the EIRB module adopts an inverse residual error network structure and comprises two branch networks, wherein an upper branch network keeps an input characteristic D unchanged, and a lower branch network is used for extracting and selecting characteristics of a small underwater object of interest and adding the characteristics with output characteristics of the upper branch network; the lower leg network includes: an expansion layer, a channel selectable component, and a compression layer, wherein,
the expansion layer is used for expanding the input characteristic channel; the size of the convolution kernel is 1 multiplied by 1, and the number of the convolution kernels is K times of the number of the input characteristic channels;
the channel selectable component is used for selecting a channel containing important information through learning weights;
the compression layer is used for compressing the characteristic channels into the quantity consistent with the input characteristics.
Preferably, the processing procedure of the EIRB module includes:
input features D ε Φ H×H×M Respectively entering two branch networks, wherein H multiplied by H is the size of an input feature, M is the number of channels of the input feature, and phi is the feature size;
for the lower branch network, the input feature D passes through the expansion layer F ex Post feature D ex The method comprises the following steps:
D ex ＝F ex (D)
D ex output feature D after entering channel selectable component se The method comprises the following steps:
D se ＝s·D ex
s＝f h (f 3 (P g (D ex )))
wherein D is se ∈Φ H×H×(K×M) S is the selection coefficient of the channel, s ε Φ 1×(K×M) ；P g For global pooling function, Φ 1×(K×M) To output feature dimensions, f 3 One-dimensional convolution layer with convolution kernel size 3, f h For the hard swish to activate the function,
from compressed layer pair D se Channel compression is carried out to obtain a characteristic D' after channel compression, and the output characteristic dimension is phi H×H×M ：
D'＝F sq (D se )
Adding the input characteristic D passing through the upper branch network and the D' output by the lower branch network to obtain an output characteristicThe method comprises the following steps:
in the method, in the process of the invention,
preferably, the method further comprises a training step of classifying the recognition model, specifically comprising:
building a training set;
and sequentially inputting the training set data into the improved MobileNet V3 network for model training, and obtaining a trained classification recognition model when the training requirement is met.
Preferably, the building a training set specifically includes:
collecting original interesting small target echo data from a real underwater environment by using a synthetic aperture sonar;
transforming the acquired echo data into voiceprint features;
the voiceprint features are randomly divided into training and testing sets according to a standard dataset format.
On the other hand, the invention provides a synthetic aperture sonar image interesting small target echo data attribute identification system, which comprises:
the target detection module is used for processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
the echo correlation module is used for correlating the position information of the small object of interest with echo data to obtain the echo data of the small object of interest;
the voiceprint feature conversion module is used for converting echo data of the small object of interest into voiceprint features; and
the attribute analysis module is used for inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of the sonar image;
the classification recognition model employs a modified MobileNet V3 network.
Preferably, the classification recognition model is deployed on an edge computing platform.
Compared with the prior art, the invention has the advantages that:
1. the invention combines the detection result of the synthetic aperture sonar image target with the classification and identification result of the echo voiceprint feature, and provides an intelligent analysis model of the underwater small object of interest, which solves the problem of low accuracy of the identification of the underwater small object of interest in the existing method in a data driving mode;
2. according to the invention, attribute analysis is carried out on the underwater small object of interest in the echo domain, so that the interpretability of the underwater small object of interest is improved, and an effective solution is provided for the underwater object fine recognition task based on the SAS image.
Drawings
FIG. 1 is a frame of the method and system for identifying the echo attribute of a small object of interest under the water of the synthetic aperture sonar;
FIG. 2 is a voiceprint feature transition flowchart;
FIG. 3 is a voiceprint feature diagram illustration;
FIG. 4 is a schematic diagram of a modified feature extraction module architecture;
fig. 5 is a flow chart of the method of the present invention.
Detailed Description
The invention provides a method for identifying the attribute of echo data of a small object of interest of a synthetic aperture sonar image, which is an underwater object analysis method based on echo domain data, and improves the adaptability of a network to the small object of interest under water by improving a feature extraction unit.
In order to achieve the above object, the present invention provides a method and a system for identifying underwater interesting small target echo data attribute based on improved MobileNet V3, wherein the method comprises:
processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
correlating the position information of the small object of interest with the echo data to obtain the echo data of the small object of interest;
transforming echo data of a small object of interest into voiceprint features;
inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of a sonar image;
the classification recognition model employs a modified MobileNet V3 network.
The system comprises: the system comprises an echo extraction module, a data set making module, a model training module and a platform deployment module;
the echo extraction module is used for capturing underwater small-object echo data of interest;
the data set making module is used for collecting underwater scene data, carrying out voiceprint feature transformation on underwater interesting small target echo data and making a data set;
the model training module is used for initializing, training and testing parameters of the attribute identification model;
the platform deployment module is used for deploying the trained attribute identification model to the embedded platform and is used for real-time online small-object-of-interest attribute identification tasks.
The synthetic aperture sonar image preprocessing module further comprises: the system comprises a synthetic aperture sonar submodule, a target detection submodule and an echo data association submodule;
the synthetic aperture sonar submodule is used for processing the received array element data to obtain a real-time synthetic aperture sonar image;
the target detection submodule is used for automatically detecting a small target of interest in the synthetic aperture sonar image to obtain the position information of the small target of interest;
the echo data association sub-module is used for associating the position information of the small object of interest in the SAS image with echo data to obtain accurate underwater small object echo data of interest;
optionally, the data set making module submodule further includes: the device comprises a data acquisition sub-module, a data cleaning sub-module and an echo data set manufacturing sub-module;
the data acquisition submodule acquires the interested small target echo data of the synthetic aperture sonar image from the real environment;
the voiceprint feature sub-module converts the interesting small target echo data into voiceprint features;
and the voiceprint data set making submodule randomly divides data into a training set and a testing set according to a standard data set format.
Optionally, the model training submodule further includes: the system comprises a parameter setting module, an improved MobileNet V3 network module and a model test module.
The parameter setting sub-module is used for completing the parameter initialization work required by model training;
the improved MobileNet V3 network sub-module is used for realizing attribute identification of the interesting small target echo data;
the model test module sub-module is used for monitoring the model training state in real time.
Optionally, the platform deployment submodule further includes: the system comprises a voiceprint feature sub-module, a model deployment sub-module and a result output sub-module;
the voiceprint feature sub-module converts the interesting small target echo data into voiceprint features;
the model deployment sub-module is used for transplanting the trained model to an edge computing platform;
and the result output sub-module is used for displaying and outputting the identification result of the small object attribute of interest.
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and examples.
Example 1
The embodiment 1 of the invention provides a method for identifying the data attribute of a small target echo of interest of a synthetic aperture sonar image, which comprises echo extraction, data set production, model training and platform deployment.
Firstly, collecting submarine small-object-of-interest data by using a synthetic aperture sonar, performing voiceprint transformation on echo data, and generating a target attribute analysis data set; secondly, initializing training parameters, training an improved MobileNet V3 model, and performing quality evaluation on a training result; thirdly, deploying the trained target detection model to an edge computing platform, and building a small target attribute distinguishing method of interest, so as to realize online detection of the small target of interest under water and output of an attribute distinguishing result. The general flow chart is shown in fig. 1, and the specific steps are as follows:
step 1, manufacturing underwater small object attribute discrimination data set
Step 1-1, acquiring original interesting small target echo data from a real underwater environment by utilizing a synthetic aperture sonar acquisition sub-module;
step 1-2, performing voiceprint feature transformation on echo data by using a playback program, as shown in fig. 2; first, processing such as sampling and quantization is performed on echo data. The preprocessed echo data is then converted into a spectral signal by a Fast Fourier Transform (FFT) or other spectral analysis technique. During the spectral analysis, the echo data is decomposed into a number of sinusoidal components of different frequencies, each component having a corresponding amplitude and phase. Next, the spectral signal is converted into a two-dimensional image of time and frequency, i.e. a voiceprint signature, as shown in fig. 3. In the voiceprint feature diagram, time is represented as a horizontal axis, frequency is represented as a vertical axis, and brightness at each time and frequency position represents the magnitude of the frequency component.
Step 1-3, a training sample set and a test sample set are established.
Step 2, model training
And 2-1, setting model training initialization parameters including batchsize, epoch, momentum, learning _rate, validization_epochs and the like in environments required by building a training platform on a deep learning server, including open source software Anaconda, pytorch, tonchvision and the like.
Step 2-2, building an improved feature extraction module (EIRB, efficient Inverted Residual Block), as shown in fig. 4. The EIRB module adopts an inverse residual error network structure, namely, adopts a strategy of expanding and compressing channels firstly, and consists of an expansion layer, a channel selectable component and a compression layer, wherein the expansion layer is responsible for expanding input characteristic channels; the channel selectable component selects a channel containing important information through learning weights; the compression layer is responsible for compressing the feature channels to a number consistent with the input features.
For an arbitrary input feature D ε Φ H×H×M Where H is the size of the input feature and M is the number of channels of the input feature. The input feature D enters two branch networks of the EIRB module: the lower side branch is responsible for extracting and selecting the characteristics of the underwater small object of interest; the upper leg keeps the input characteristic D unchanged and finally adds to the output characteristic of the top leg network. For the lower branch network, the input feature D first passes through the expansion layer, and the mathematical expression of the output feature is:
D ex ＝F ex (D),D∈Φ H×H×M (1)
wherein D is an original input feature; d (D) ex For features after passing through the expansion layer, the convolution kernel size of the expansion layer is 1×1, and the number of convolution kernels is K times that of the input feature channels, namely k×m.
Then, output the feature D ex The mathematical expression of the output characteristics of the input ECA channel selection component is as follows:
D se ＝s·D ex (2)
s＝f h (f 3 (P g (D ex ))) (3)
wherein D is se Channel characteristics after channel selection; s is the selection coefficient of the channel, s epsilon phi 1×(K×M) ；P g () For global pooling function, the output characteristic dimension is phi 1×(K×M) ；f 3 For a one-dimensional convolution layer with a convolution kernel size of 3, the output characteristic dimension is phi 1×(K×M) ；f h For the hard swish to activate the function,
then, for D se And carrying out channel compression, wherein the mathematical expression is as follows:
D'＝F sq (D se ),D se ∈Φ H×H×(K×M) (4)
wherein D' is the characteristic of the compressed channel, and the dimension of the output characteristic is phi H×H×M 。
Through the calculation, the output characteristic mathematical expression of the EIRB module can be finally obtained as follows:
in the method, in the process of the invention,for the output characteristics of EIRB module, +.>The feature map size is H×H, and the number of channels is M.
Step 2-3, an improved MobileNet V3 network is built. The modified MobileNet V3 network is shown in table 1, in which the input channel, intermediate channel, output channel, step size and activation function (RE represents the ReLU activation function, HS represents the H-Swich activation function) of each convolutional layer remain identical to the original MobileNet V3 network; except that modified mobilet V3 uses EIRB to replace IRB in original mobilet V3 + And (5) a module.
Table 1 improved MobileNet V3 network architecture
And 2-3, monitoring the training process and the test result of the MobileNet V3P network in real time, and stopping training when the evaluation index meets the requirement.
Step 3, platform deployment
And 3-1, building an attribute identification model running environment on the edge computing platform.
And 3-2, constructing an underwater small object attribute identification method, wherein the flow is shown in figure 5. Firstly, obtaining the position information of a small object of interest of a high-resolution large-size sonar image by using an object detection method; then, associating the image coordinate information of the small object of interest with the array element domain data, and extracting the corresponding array element domain data; then, converting the echo data into voiceprint features; and finally, classifying and identifying the small-object-of-interest voiceprint features of the sonar image by using the improved MobileNet V3 network, so as to realize attribute analysis of the small-object-of-interest sonar image.
And 4, displaying the identification result of the small object attribute of interest.
Simulation experiment:
the technical effects of the invention are further described below in conjunction with simulation experiments:
the experimental training hardware platform is i7-8750H, the memory is 32GB (16 GB x 2), the GPU is 2070s (8G), and the software environment is win10, python3.7, torch1.3.1, torchvision0.4.2 and the like. The size of the input image was 224 pixels, the batch size of all models was set to 16, the learning rate was a variable learning rate, and the value was 0.01. The experimental data contains 4 metal targets and 3 reef targets, the data set is denoted as MS, and the data composition is shown in Table 2.
Table 2MS dataset composition
The present experiment compares and analyzes the performance differences of mobilet V3 with the target detection method herein MobileNet V3P on the dataset MS. And respectively recording the parameter size, the operation time and the accuracy of the detection model to the MS test data set.
TABLE 3 Attribute identification model Performance comparison
As can be seen from Table 3, the classification accuracy of MobileNet V3P of the present invention is 1.04% higher than that of MobileNet V3 respectively; the operation time is reduced by 5ms; the parameters were reduced by 1.99MB. Considering comprehensively, the MobileNet V3P is more suitable for the task of identifying the attribute of the interesting small target echo data based on the synthetic aperture sonar image.
Example 2
The embodiment 2 of the invention provides a system for identifying the attribute of echo data of a small object of interest of a synthetic aperture sonar image, which is realized based on the method of the embodiment 1, and comprises the following steps:
the target detection module is used for processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
the echo correlation module is used for correlating the position information of the small object of interest with echo data to obtain the echo data of the small object of interest;
the voiceprint feature conversion module is used for converting echo data of the small object of interest into voiceprint features;
the attribute analysis module is used for inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of the sonar image;
the classification recognition model employs a modified MobileNet V3 network.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and are not limiting. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the appended claims.
Claims (8)
1. A synthetic aperture sonar image small target echo data attribute identification method, the method comprising:
processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
correlating the position information of the small object of interest with the echo data to obtain the echo data of the small object of interest;
transforming echo data of a small object of interest into voiceprint features;
inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of a sonar image;
the classification recognition model employs a modified MobileNet V3 network.
2. The method for identifying the echo data attribute of the small object of interest in the synthetic aperture sonar image according to claim 1, wherein the echo data of the small object of interest is converted into voiceprint characteristics; the method specifically comprises the following steps:
preprocessing the echo data of the small object of interest, including sampling and quantization;
performing fast Fourier transform or other spectrum analysis technology on the preprocessed echo data, and converting the echo data into a spectrum signal;
and converting the frequency spectrum signal into a two-dimensional image of time and frequency to obtain the voiceprint feature map.
3. The method for identifying the attribute of the small target echo data of interest of the synthetic aperture sonar image according to claim 1, wherein the improved mobilet V3 network adopts an EIRB module to replace IRB in the original mobilet V3 + A module; the EIRB module adopts an inverse residual error network structure and comprises two branch networks, wherein an upper branch network keeps an input characteristic D unchanged, and a lower branch network is used for extracting and selecting characteristics of a small underwater object of interest and adding the characteristics with output characteristics of the upper branch network; the lower leg network includes: an expansion layer, a channel selectable component, and a compression layer, wherein,
the expansion layer is used for expanding the input characteristic channel; the size of the convolution kernel is 1 multiplied by 1, and the number of the convolution kernels is K times of the number of the input characteristic channels;
the channel selectable component is used for selecting a channel containing important information through learning weights;
the compression layer is used for compressing the characteristic channels into the quantity consistent with the input characteristics.
4. A method for identifying the attribute of echo data of a small object of interest in a synthetic aperture sonar image according to claim 3, wherein the processing procedure of the EIRB module includes:
input features D ε Φ H×H×M Respectively entering two branch networks, wherein H multiplied by H is the size of an input feature, M is the number of channels of the input feature, and phi is the feature size;
for the lower branch network, the input feature D passes through the expansion layer F ex Post feature D ex The method comprises the following steps:
D ex ＝F ex (D)
D ex output feature D after entering channel selectable component se The method comprises the following steps:
D se ＝s·D ex
s＝f h (f 3 (P g (D ex )))
wherein D is se ∈Φ H×H×(K×M) S is the selection coefficient of the channel, s ε Φ 1×(K×M) ；P g For global pooling function, Φ 1×(K×M) To output feature dimensions, f 3 One-dimensional convolution layer with convolution kernel size 3, f h For the hard swish to activate the function,
from compressed layer pair D se Channel compression is carried out to obtain a characteristic D' after channel compression, and the output characteristic dimension is phi H×H×M ：
D'＝F sq (D se )
Adding the input characteristic D passing through the upper branch network and the D' output by the lower branch network to obtain an output characteristic D as follows:
in the method, in the process of the invention,
5. the method for identifying the attribute of the echo data of the small object of interest of the synthetic aperture sonar image according to claim 1, wherein said method further comprises a training step of classifying and identifying the model, and specifically comprises the following steps:
building a training set;
and sequentially inputting the training set data into the improved MobileNet V3 network for model training, and obtaining a trained classification recognition model when the training requirement is met.
6. The method for identifying the attribute of the echo data of the small object of interest in the synthetic aperture sonar image according to claim 5, wherein said creating the training set specifically comprises:
collecting original interesting small target echo data from a real underwater environment by using a synthetic aperture sonar;
transforming the acquired echo data into voiceprint features;
the voiceprint features are randomly divided into training and testing sets according to a standard dataset format.
7. A synthetic aperture sonar image small target echo data attribute identification system, the system comprising:
the target detection module is used for processing the received echo data of the synthetic aperture sonar to obtain a real-time synthetic aperture sonar image, and obtaining the position information of the small target of interest by adopting a target detection method;
the echo correlation module is used for correlating the position information of the small object of interest with echo data to obtain the echo data of the small object of interest;
the voiceprint feature conversion module is used for converting echo data of the small object of interest into voiceprint features; and
the attribute analysis module is used for inputting voiceprint features into a pre-established and trained classification recognition model to realize attribute analysis of small objects of interest of the sonar image;
the classification recognition model employs a modified MobileNet V3 network.
8. The synthetic aperture sonar image small target echo data attribute of interest recognition system of claim 7, wherein the classification recognition model is deployed on an edge computing platform.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311321576.3A CN117434524A (en) | 2023-10-12 | 2023-10-12 | Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311321576.3A CN117434524A (en) | 2023-10-12 | 2023-10-12 | Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117434524A true CN117434524A (en) | 2024-01-23 |
Family
ID=89549002
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311321576.3A Pending CN117434524A (en) | 2023-10-12 | 2023-10-12 | Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117434524A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2740158C1 (en) * | 2020-05-25 | 2021-01-12 | Акционерное Общество "Концерн "Океанприбор" | Method for sonar detection of high-speed small-size object |
| CN112613504A (en) * | 2020-12-17 | 2021-04-06 | 上海大学 | Sonar underwater target detection method |
| CN113589300A (en) * | 2021-06-29 | 2021-11-02 | 中国船舶重工集团公司第七一五研究所 | Synthetic aperture sonar submerged target imaging enhancement method based on compressed sensing |
| CN115223033A (en) * | 2022-08-29 | 2022-10-21 | 中国科学院声学研究所 | Synthetic aperture sonar image target classification method and system |
| CN115410079A (en) * | 2022-09-02 | 2022-11-29 | 河南大学 | Synthetic aperture sonar image target detection method based on impulse neural network |
| CN116188963A (en) * | 2022-11-16 | 2023-05-30 | 宜昌测试技术研究所 | Unmanned ship target detection and autonomous identification system and method based on deep learning |
| CN116343016A (en) * | 2023-01-03 | 2023-06-27 | 中国科学院声学研究所 | Multi-angle sonar image target classification method based on lightweight convolution network |
-
2023
- 2023-10-12 CN CN202311321576.3A patent/CN117434524A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2740158C1 (en) * | 2020-05-25 | 2021-01-12 | Акционерное Общество "Концерн "Океанприбор" | Method for sonar detection of high-speed small-size object |
| CN112613504A (en) * | 2020-12-17 | 2021-04-06 | 上海大学 | Sonar underwater target detection method |
| CN113589300A (en) * | 2021-06-29 | 2021-11-02 | 中国船舶重工集团公司第七一五研究所 | Synthetic aperture sonar submerged target imaging enhancement method based on compressed sensing |
| CN115223033A (en) * | 2022-08-29 | 2022-10-21 | 中国科学院声学研究所 | Synthetic aperture sonar image target classification method and system |
| CN115410079A (en) * | 2022-09-02 | 2022-11-29 | 河南大学 | Synthetic aperture sonar image target detection method based on impulse neural network |
| CN116188963A (en) * | 2022-11-16 | 2023-05-30 | 宜昌测试技术研究所 | Unmanned ship target detection and autonomous identification system and method based on deep learning |
| CN116343016A (en) * | 2023-01-03 | 2023-06-27 | 中国科学院声学研究所 | Multi-angle sonar image target classification method based on lightweight convolution network |
Non-Patent Citations (2)
| Title |
|---|
| 李宝奇;黄海宁;刘纪元;刘正君;韦琳哲: "基于改进SSD的合成孔径声呐图像水下多尺度目标轻量化检测模型", 电子与信息学报, no. 010, 31 December 2021 (2021-12-31) * |
| 李宝奇;黄海宁;刘纪元;李宇: "基于改进CycleGAN的光学图像迁移生成水下小目标合成孔径声纳图像算法研究", 电子学报, no. 009, 31 December 2021 (2021-12-31) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245608B (en) | Underwater target identification method based on half tensor product neural network | |
| CN109446902A (en) | A kind of marine environment based on unmanned platform and the comprehensive cognitive method of target | |
| US20210073611A1 (en) | Dynamic data structures for data-driven modeling | |
| CN108520758B (en) | Visual-auditory cross-modal object material retrieval method and system | |
| CN108805056B (en) | Camera monitoring face sample expansion method based on 3D face model | |
| Sun et al. | Underwater single-channel acoustic signal multitarget recognition using convolutional neural networks | |
| Zhou et al. | A denoising representation framework for underwater acoustic signal recognition | |
| CN111856422A (en) | Lip language identification method based on broadband multichannel millimeter wave radar | |
| Wang et al. | Audio event detection and classification using extended R-FCN approach | |
| CN112289326A (en) | Bird identification comprehensive management system with noise removal function and noise removal method thereof | |
| CN115661649A (en) | Ship-borne microwave radar image oil spill detection method and system based on BP neural network | |
| Xu et al. | Self-supervised learning–based underwater acoustical signal classification via mask modeling | |
| CN112101440B (en) | Coronage substance ejection detection method based on improved U-Net network | |
| CN110503157B (en) | Image steganalysis method of multitask convolution neural network based on fine-grained image | |
| CN117434524A (en) | Method for identifying attribute of echo data of small object of interest in synthetic aperture sonar image | |
| Sun et al. | Robust high-resolution direction-of-arrival estimation method using DenseBlock-based U-net | |
| Wu et al. | UNDERWATER ACOUSTIC SIGNAL ANALYSIS: PREPROCESSING AND CLASSIFICATION BY DEEP LEARNING. | |
| CN112735442B (en) | Wetland ecology monitoring system with audio separation voiceprint recognition function and audio separation method thereof | |
| Siva Teja et al. | Identification of sloshing noises using convolutional neural network | |
| CN107341798A (en) | High Resolution SAR image change detection method based on global local SPP Net | |
| Ashraf et al. | Improved CycleGAN for underwater ship engine audio translation | |
| CN109344767B (en) | SAR target identification method based on multi-azimuth multi-feature collaborative representation | |
| Chen et al. | Segmentation of sonar image on seafloor sediments based on multiclass SVM | |
| Hu et al. | A generalized network based on multi-scale densely connection and residual attention for sound source localization and detection | |
| CN112687280A (en) | Biodiversity monitoring system with frequency spectrum-time space interface |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |