CN113537020A - Complex SAR image target identification method based on improved neural network - Google Patents
Complex SAR image target identification method based on improved neural network Download PDFInfo
- Publication number
- CN113537020A CN113537020A CN202110769105.3A CN202110769105A CN113537020A CN 113537020 A CN113537020 A CN 113537020A CN 202110769105 A CN202110769105 A CN 202110769105A CN 113537020 A CN113537020 A CN 113537020A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- target recognition
- sar
- sample
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The application relates to a complex SAR image target identification method based on an improved neural network. The method comprises the following steps: the SAR image with three channels is generated by utilizing the specific complex information of the SAR image, meanwhile, the target recognition neural network constructed on the basis of the channel attention mechanism module and the residual error network can adaptively pay attention to the useful characteristics of each channel, so that the target recognition accuracy of the SAR image is improved, the target recognition neural network constructed on the basis of the three-channel SAR image is trained, the trained neural network can effectively utilize the complex information of the SAR image and extract the characteristics related to the target, and the target recognition accuracy of the SAR image is improved.
Description
Technical Field
The application relates to the technical field of radar image processing, in particular to a complex SAR image target identification method based on an improved neural network.
Background
Synthetic Aperture Radar (SAR) adopts microwave coherent imaging, so that an SAR image is complex in nature, and the SAR image is richer in pixel information quantity, which is also obviously different from a common optical infrared remote sensing image. The method can acquire two-dimensional high-resolution images all day long and all weather, and is an important means for monitoring marine targets in all countries in the world at present.
The current SAR image ship target identification method can be summarized as 1) an identification method based on effective feature extraction. The target characteristics in the image can be described from different angles by extracting effective features, and then an effective classification method is selected for target identification. The current feature expression method is mainly divided into spatial features, statistical features, transform domain features and algebraic features. 2) An identification method based on feature fusion. The feature fusion can increase the feature information of the image and realize more comprehensive feature expression. 3) And (3) a recognition method based on model matching. And obtaining model prediction characteristics according to the training images, and then realizing target identification by matching the prediction characteristics of the models and the extraction characteristics of the images. 4) An identification method based on SAR imaging principle and ground object electromagnetic scattering mechanism. And the classification and identification precision is improved by using the characteristics of strong scattering information of the target, orientation invariance and the like in SAR data. 5) An identification method based on deep learning. With the rapid development of the deep learning method, the method is widely applied to the field of target recognition, the accuracy of target recognition is greatly improved by establishing the relationship between the low-level features and the high-level semantics through the machine learning features of the unsupervised or supervised learning method, and the great advantages and potentials of the deep learning in the target recognition are fully embodied.
In recent years, deep learning frameworks such as CNN and FCN are used for SAR image target recognition, and good results are obtained. However, in the current deep learning framework applied to the SAR image recognition, for the amplitude information, the complex SAR image needs to be projected into a real number domain represented by the amplitude, and the complex information specific to the SAR image cannot be effectively utilized.
Disclosure of Invention
In view of the above, there is a need to provide a complex SAR image target identification method based on an improved neural network, which can effectively improve the accuracy of SAR image target identification.
A complex SAR image target identification method based on an improved neural network, the method comprises the following steps:
acquiring a plurality of complex SAR sample images, respectively extracting a real part image, an imaginary part image and an amplitude image of each complex SAR sample image, recombining to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and constructing a sample data set according to each three-channel SAR sample image;
carrying out standardization processing on the sample data set to obtain a data mean value and a data standard deviation of the sample data set;
constructing a target recognition neural network combining a channel attention mechanism module and a residual error network, setting parameters of the target recognition neural network according to the data mean value and the data standard deviation, and inputting a sample data set into the target recognition neural network for training to obtain a trained target recognition neural network;
acquiring a complex SAR image to be identified, extracting a real part image, an imaginary part image and an amplitude image of the complex SAR image, and fusing to obtain a three-channel SAR image corresponding to the complex SAR image;
and inputting the three-channel SAR image into a trained target recognition neural network for target recognition.
In one embodiment, the original gray scale is preserved when the real part image, the imaginary part image and the amplitude image of the complex SAR sample image are recombined.
In one embodiment, the sample data set is normalized by a standard score method.
In one embodiment, the constructing the target recognition neural network of the channel attention mechanism module in combination with the residual error network includes:
connecting the channel attention mechanism module to each residual learning unit "identity" branch in a residual network to construct the target recognition neural network.
In one embodiment, label smoothing regularization is further added to train the target recognition neural network to constrain the target recognition neural network, so that a loss function is:
wherein ∈ is a hyper-parameter, K is a category number of each plural SAR sample image in the sample data set, q (K) is a label distribution, p (K) is a prediction distribution, K is a certain category label of each plural SAR sample image in the sample data set, and y is a true label.
According to the complex SAR image target recognition method based on the improved neural network, the three-channel SAR image is generated by utilizing the specific complex information of the SAR image, meanwhile, the target recognition neural network constructed based on the channel attention mechanism module and the residual error network can adaptively pay attention to the useful characteristics of each channel, so that the SAR image target recognition accuracy is improved, the three-channel SAR image is utilized to train the constructed target recognition neural network, the trained neural network can effectively utilize the complex information of the SAR image and extract the characteristics related to the target, and the SAR image target recognition accuracy is improved.
Drawings
FIG. 1 is a schematic flow chart of a complex SAR image target identification method based on an improved neural network in one embodiment;
FIG. 2 is a schematic diagram of complex SAR image data enhancement in one embodiment;
FIG. 3 is a diagram of a target recognition neural network architecture in one embodiment;
FIG. 4 is a schematic diagram of a channel attention mechanism module in one embodiment;
FIG. 5 is a graph comparing experimental results for different target recognition methods in one embodiment;
FIG. 6 is a block diagram of a complex SAR image target recognition device based on an improved neural network in one embodiment;
FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
As shown in fig. 1, a complex SAR image target recognition method based on an improved neural network is provided, which includes the following steps:
step S100, obtaining a plurality of complex SAR sample images, respectively extracting a real part image, an imaginary part image and an amplitude image of each complex SAR sample image, recombining to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and constructing a sample data set according to each three-channel SAR sample image;
step S110, carrying out standardization processing on the sample data set to obtain a data mean value and a data standard deviation of the sample data set;
step S120, a target recognition neural network combining a channel attention mechanism module and a residual error network is constructed, parameters of the target recognition neural network are set according to the data mean value and the data standard deviation, and then a sample data set is input into the target recognition neural network for training to obtain a trained target recognition neural network;
step S130, acquiring a complex SAR image to be identified, extracting a real part image, an imaginary part image and an amplitude image of the complex SAR image, and fusing to obtain a three-channel SAR image corresponding to the complex SAR image;
and step S140, inputting the three-channel SAR image into the trained target recognition neural network for target recognition.
In order to solve the problem that the SAR image has the characteristic of complex information, and when the existing deep learning network is used for target identification of the SAR image, the complex SAR image is required to be projected into a real number domain represented by amplitude, namely, a method for effectively identifying the target by using the complex information of the SAR image is not provided when the deep learning network is used for identification.
In the embodiment, the target recognition of the SAR complex image is realized based on an improved neural network.
In step S100, a sample data set for training the improved neural network is constructed. A plurality of complex SAR sample images are first acquired. The content of these complex SAR sample images includes multiple categories of images of the same target. For example, including SAR images of different types of vehicles, there may be multiple different SAR images for one type of vehicle.
Then, the real part image, the imaginary part image and the amplitude image of each complex SAR sample image are extracted, and the three images are recombined to generate a three-channel SAR image, that is, the complex SAR sample image is enhanced, as shown in fig. 2.
Specifically, the complex SAR sample image may be represented in a complex form as Pixel r + i · j Aejφ，
Where r is the real part, i is the imaginary part, j is the imaginary unit and has j2A is amplitude (i.e., gray scale information). In complex SAR imageThe real part r, the imaginary part i and the amplitude A are combined into a three-channel SAR image, which can be expressed as Z [ | | | | | | i | A [ | r | | | i | A]。
In this embodiment, in the process of reconstructing the real part image, the imaginary part image and the amplitude image of the complex SAR sample image, in order to reduce the loss of data pixel information, the pixel value of the data is not scaled to the interval of 0-255, but is retained as the original 32-bit gray scale.
In step S110, a standard score method is used to perform normalization processing on the sample data set.
Specifically, the normalization process is represented as:
in formula (1), S is the standard deviation of the three-channel image data in the data sample set,
is the mean value of three-channel image data in a data sample set, N is the total pixel amount of the data sample set, xiAre three channel image data pixel values in a data sample set.
In the implementation, the parameters of the constructed target identification network are set according to the data mean value and the data standardization, so that on one hand, the precision of a network model can be improved, and the accuracy of the classifier can be improved. On the other hand, the solving speed of gradient descent can be increased, and the convergence of the network model is accelerated.
And before inputting the sample data set into a target recognition neural network for training, labeling each three-channel image according to the category in the image.
In step S120, constructing the target recognition neural network in which the channel attention mechanism module is combined with the residual error network includes: a channel attention mechanism module is connected to each residual learning unit "identity" branch in the residual network to construct a target recognition neural network.
Specifically, a Squeeze-and-excitation (SE) channel attention mechanism module is introduced on the basis of a residual error network (Resnet18 network) to improve the sensitivity of the network to channel characteristics. The SE module is added to the "identity" branch of the residue blocks of the Resnet18 network, so that a weight prediction branch is added to each residue block, so that the network can pay more attention to the features of the objects in the image when learning the image features, and the structure of the object recognition neural network is shown in fig. 3.
Specifically, the Resnet18 network includes a convolutional layer, a pooling layer, four residual learning units connected in sequence, an average pooling layer, and a fully-connected layer.
When the data is processed through the residual block, the input data is processed through the convolution layer, the batch normalization layer, the ReLu activation layer, the other convolution layer and the other batch normalization layer in sequence to obtain intermediate data, the input data is subjected to residual processing with the intermediate data through the identity branch and then subjected to ReLu activation processing to obtain data output of the residual block. Wherein the SE block adds an identity branch to add a weighted prediction branch to each residual block.
In one implementation, the SE module includes a global pooling layer, a full connection layer, a ReLu activation layer, a full connection layer, and a Sigmoid activation layer, which are connected in sequence, as shown in fig. 4.
In the process of training the target recognition neural network, in order to solve the problem that samples in a sample data set are few, label smoothing regularization is added to constrain the target recognition neural network so as to reduce the phenomenon of overfitting, and a hyper-parameter is added so that a loss function is as follows:
wherein ∈ is a hyper-parameter, K is a category number of each plural SAR sample image in the sample data set, q (K) is a label distribution, p (K) is a prediction distribution, K is a certain category label of each plural SAR sample image in the sample data set, and y is a true label.
The smooth regularization of the labels is equivalent to reducing the weight of the category of the real sample labels when loss is calculated, the probability of error labels is not 0, and finally the overfitting inhibition effect is achieved.
When the target recognition neural network is actually trained, the complex SAR sample images in each category in the sample data set are randomly divided into a training set and a testing set, and the division ratio is 9: 1. Firstly, the training set is used for carrying out initial training on the target recognition neural network, and after the training is finished, the training set is used for carrying out fine tuning on the trained neural network so that the accuracy of the target recognition neural network can reach a preset target.
In steps S130-S140, the trained target recognition neural network is used, and first a real part image, an imaginary part image, and an amplitude image of the complex SAR image to be recognized need to be extracted and fused to obtain a three-channel SAR image corresponding to the complex SAR image, and then the three-channel SAR image is input into the trained target recognition neural network for target recognition.
As shown in fig. 5, (a) and (b) of fig. 5 show that the recognition result using the three-channel SAR image on the Resnet18 network is improved by 1% -2% compared with the recognition result using the single-channel SAR image, which indicates that the local representation provided by the complex information can be added on the basis of the amplitude information on the basis of forming the real part and imaginary part information in the complex information into the three-channel SAR image, so that the recognition effect is improved.
As can be seen from (b) and (c) of fig. 5, when the three-channel SAR image is used in the method and the Resnet18 network, the recognition effect is improved by 1.5%. The channel attention mechanism of the method can realize the self-adaptive learning of the important characteristics of each channel and inhibit background information, so that the method is superior to other methods.
In the complex SAR image target identification method based on the improved neural network, the data of the complex SAR image is enhanced, the real part, the imaginary part and the amplitude in the complex SAR image are recombined into a three-channel SAR image, complex information is introduced to enhance the details of local representation, and then the three-channel SAR image is trained on a target identification neural network based on the combination of a channel attention mechanism module and a residual error network, wherein the channel attention mechanism is combined with a Resnet18 network, so that the network can adaptively pay attention to the useful characteristics of each channel, and the accuracy of SAR image target identification is improved. The method can effectively utilize the complex information in the SAR image, so that the data enhances the detail of local representation on the basis of the original amplitude information, and the method is simple and easy to realize.
It should be understood that, although the steps in the flowchart of fig. 1 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 1 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
In one embodiment, as shown in fig. 6, there is provided an improved neural network-based complex SAR image target recognition apparatus, including: the system comprises a sample data set construction module 200, a standardization processing module 210, a neural network training module 220, a three-channel SAR image obtaining module 230 and a target identification module 240, wherein:
the sample data set construction module 200 is configured to acquire a plurality of complex SAR sample images, extract a real part image, an imaginary part image, and an amplitude image of each complex SAR sample image, and perform recombination to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and construct a sample data set according to each three-channel SAR sample image;
the normalization processing module 210 is configured to perform normalization processing on the sample data set to obtain a data mean and a data standard deviation of the sample data set;
the neural network training module 220 is used for constructing a target recognition neural network combining the channel attention mechanism module and the residual error network, setting parameters of the target recognition neural network according to the data mean value and the data standard deviation, and inputting a sample data set into the target recognition neural network for training to obtain a trained target recognition neural network;
a three-channel SAR image obtaining module 230, configured to obtain a complex SAR image to be identified, extract a real part image, an imaginary part image, and an amplitude image of the complex SAR image, and perform fusion to obtain a three-channel SAR image corresponding to the complex SAR image;
and a target identification module 240, configured to input the three-channel SAR image into a trained target identification neural network for target identification.
For specific limitations of the complex SAR image target recognition device based on the improved neural network, reference may be made to the above limitations of the complex SAR image target recognition method based on the improved neural network, and details are not repeated here. The modules in the complex SAR image target recognition device based on the improved neural network can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 7. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a complex SAR image target recognition method based on an improved neural network. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:
acquiring a plurality of complex SAR sample images, respectively extracting a real part image, an imaginary part image and an amplitude image of each complex SAR sample image, recombining to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and constructing a sample data set according to each three-channel SAR sample image;
carrying out standardization processing on the sample data set to obtain a data mean value and a data standard deviation of the sample data set;
constructing a target recognition neural network combining a channel attention mechanism module and a residual error network, setting parameters of the target recognition neural network according to the data mean value and the data standard deviation, and inputting a sample data set into the target recognition neural network for training to obtain a trained target recognition neural network;
acquiring a complex SAR image to be identified, extracting a real part image, an imaginary part image and an amplitude image of the complex SAR image, and fusing to obtain a three-channel SAR image corresponding to the complex SAR image;
and inputting the three-channel SAR image into a trained target recognition neural network for target recognition.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
and reserving original gray scale when the real part image, the imaginary part image and the amplitude image of the complex SAR sample image are recombined.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
and carrying out standardization processing on the sample data set by adopting a standard score method.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
the target recognition neural network combining the channel attention mechanism building module and the residual error network comprises:
connecting the channel attention mechanism module to each residual learning unit "identity" branch in a residual network to construct the target recognition neural network.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
and when the target recognition neural network is trained, label smoothing regularization is added to constrain the target recognition neural network, so that a loss function is as follows:
wherein ∈ is a hyper-parameter, K is a category number of each plural SAR sample image in the sample data set, q (K) is a label distribution, p (K) is a prediction distribution, K is a certain category label of each plural SAR sample image in the sample data set, and y is a true label.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:
acquiring a plurality of complex SAR sample images, respectively extracting a real part image, an imaginary part image and an amplitude image of each complex SAR sample image, recombining to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and constructing a sample data set according to each three-channel SAR sample image;
carrying out standardization processing on the sample data set to obtain a data mean value and a data standard deviation of the sample data set;
constructing a target recognition neural network combining a channel attention mechanism module and a residual error network, setting parameters of the target recognition neural network according to the data mean value and the data standard deviation, and inputting a sample data set into the target recognition neural network for training to obtain a trained target recognition neural network;
acquiring a complex SAR image to be identified, extracting a real part image, an imaginary part image and an amplitude image of the complex SAR image, and fusing to obtain a three-channel SAR image corresponding to the complex SAR image;
and inputting the three-channel SAR image into a trained target recognition neural network for target recognition.
In one embodiment, the computer program when executed by the processor further performs the steps of:
and reserving original gray scale when the real part image, the imaginary part image and the amplitude image of the complex SAR sample image are recombined.
In one embodiment, the computer program when executed by the processor further performs the steps of:
and carrying out standardization processing on the sample data set by adopting a standard score method.
In one embodiment, the computer program when executed by the processor further performs the steps of:
the target recognition neural network combining the channel attention mechanism building module and the residual error network comprises:
connecting the channel attention mechanism module to each residual learning unit "identity" branch in a residual network to construct the target recognition neural network.
In one embodiment, the computer program when executed by the processor further performs the steps of:
and when the target recognition neural network is trained, label smoothing regularization is added to constrain the target recognition neural network, so that a loss function is as follows:
wherein ∈ is a hyper-parameter, K is a category number of each plural SAR sample image in the sample data set, q (K) is a label distribution, p (K) is a prediction distribution, K is a certain category label of each plural SAR sample image in the sample data set, and y is a true label.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (5)
1. The method for recognizing the target of the complex SAR image based on the improved neural network is characterized by comprising the following steps of:
acquiring a plurality of complex SAR sample images, respectively extracting a real part channel image, an imaginary part image and an amplitude image of each complex SAR sample image, recombining to obtain a three-channel SAR sample image corresponding to each complex SAR sample image, and constructing a sample data set according to each three-channel SAR sample image;
carrying out standardization processing on the sample data set to obtain a data mean value and a data standard deviation of the sample data set;
constructing a target recognition neural network combining a channel attention mechanism module and a residual error network, setting parameters of the target recognition neural network according to the data mean value and the data standard deviation, and inputting a sample data set into the target recognition neural network for training to obtain a trained target recognition neural network;
acquiring a complex SAR image to be identified, extracting a real part image, an imaginary part image and an amplitude image of the complex SAR image, and fusing to obtain a three-channel SAR image corresponding to the complex SAR image;
and inputting the three-channel SAR image into a trained target recognition neural network for target recognition.
2. The complex SAR image target identification method in claim 1, characterized in that the original gray scale is preserved when the real part image, imaginary part image and amplitude image of the complex SAR sample image are recombined.
3. The method for complex SAR image target recognition according to claim 1, characterized in that a standard score method is adopted to perform standardization processing on the sample data set.
4. The method for target recognition of a complex SAR image as claimed in claim 1, wherein the constructing of the target recognition neural network in which the channel attention mechanism module is combined with the residual error network comprises:
connecting the channel attention mechanism module to each residual learning unit "identity" branch in a residual network to construct the target recognition neural network.
5. The method for complex SAR image target recognition according to claim 1, characterized in that a label smoothing regularization is added to constrain the target recognition neural network when the target recognition neural network is trained, so that a loss function is:
wherein ∈ is a hyper-parameter, K is a category number of each plural SAR sample image in the sample data set, q (K) is a label distribution, p (K) is a prediction distribution, K is a certain category label of each plural SAR sample image in the sample data set, and y is a true label.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110769105.3A CN113537020B (en) | 2021-07-07 | 2021-07-07 | Complex SAR image target identification method based on improved neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110769105.3A CN113537020B (en) | 2021-07-07 | 2021-07-07 | Complex SAR image target identification method based on improved neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113537020A true CN113537020A (en) | 2021-10-22 |
| CN113537020B CN113537020B (en) | 2022-12-09 |
Family
ID=78127037
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110769105.3A Active CN113537020B (en) | 2021-07-07 | 2021-07-07 | Complex SAR image target identification method based on improved neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537020B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114581766A (en) * | 2022-01-19 | 2022-06-03 | 南京航空航天大学 | Target classification method based on three-channel sparse SAR image data |
| CN115223398A (en) * | 2022-07-14 | 2022-10-21 | 集美大学 | Nuclear adaptive fractional order complex value neural network AIS interpolation method considering channel constraint |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN109145815A (en) * | 2018-08-21 | 2019-01-04 | 深圳大学 | A kind of SAR target identification method, device, computer equipment and storage medium |
| CN110705508A (en) * | 2019-10-15 | 2020-01-17 | 中国人民解放军战略支援部队航天工程大学 | Satellite identification method of ISAR image |
| CN110969121A (en) * | 2019-11-29 | 2020-04-07 | 长沙理工大学 | High-resolution radar target recognition algorithm based on deep learning |
| CN111126570A (en) * | 2019-12-24 | 2020-05-08 | 江西理工大学 | SAR target classification method for pre-training complex number full convolution neural network |
| CN111339827A (en) * | 2020-01-18 | 2020-06-26 | 中国海洋大学 | SAR image change detection method based on multi-region convolutional neural network |
| CN111562578A (en) * | 2020-03-05 | 2020-08-21 | 华北水利水电大学 | Distributed array SAR sparse representation three-dimensional imaging algorithm considering scene amplitude real value constraint |
| CN112419198A (en) * | 2020-11-27 | 2021-02-26 | 中国矿业大学 | Non-local mean weighting method for SAR interferogram filtering |
| CN113011305A (en) * | 2021-03-12 | 2021-06-22 | 中国人民解放军国防科技大学 | SAR image road extraction method and device based on semantic segmentation and conditional random field |
| CN113065433A (en) * | 2021-03-23 | 2021-07-02 | 南京航空航天大学 | SAR target recognition method based on sparse complex image |
-
2021
- 2021-07-07 CN CN202110769105.3A patent/CN113537020B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN109145815A (en) * | 2018-08-21 | 2019-01-04 | 深圳大学 | A kind of SAR target identification method, device, computer equipment and storage medium |
| CN110705508A (en) * | 2019-10-15 | 2020-01-17 | 中国人民解放军战略支援部队航天工程大学 | Satellite identification method of ISAR image |
| CN110969121A (en) * | 2019-11-29 | 2020-04-07 | 长沙理工大学 | High-resolution radar target recognition algorithm based on deep learning |
| CN111126570A (en) * | 2019-12-24 | 2020-05-08 | 江西理工大学 | SAR target classification method for pre-training complex number full convolution neural network |
| CN111339827A (en) * | 2020-01-18 | 2020-06-26 | 中国海洋大学 | SAR image change detection method based on multi-region convolutional neural network |
| CN111562578A (en) * | 2020-03-05 | 2020-08-21 | 华北水利水电大学 | Distributed array SAR sparse representation three-dimensional imaging algorithm considering scene amplitude real value constraint |
| CN112419198A (en) * | 2020-11-27 | 2021-02-26 | 中国矿业大学 | Non-local mean weighting method for SAR interferogram filtering |
| CN113011305A (en) * | 2021-03-12 | 2021-06-22 | 中国人民解放军国防科技大学 | SAR image road extraction method and device based on semantic segmentation and conditional random field |
| CN113065433A (en) * | 2021-03-23 | 2021-07-02 | 南京航空航天大学 | SAR target recognition method based on sparse complex image |
Non-Patent Citations (5)
| Title |
|---|
| PING HAN 等: "Airport runway detection based on a combination of complex convolution and ResNet for PolSAR images", 《2021 SAR IN BIG DATA ERA (BIGSARDATA)》, 26 May 2021 (2021-05-26) * |
| RONG YANG 等: "A Novel Polarimetric SAR Classification Method Integrating Pixel-Based and Patch-Based Classification", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》, vol. 17, no. 3, 4 July 2019 (2019-07-04), XP011774949, DOI: 10.1109/LGRS.2019.2923403 * |
| 孙忠镇 等: "基于级联网络的复杂大场景SAR图像舰船目标快速检测", 《信号处理》, vol. 37, no. 6, 25 March 2021 (2021-03-25) * |
| 李广帅 等: "基于卷积神经网络与注意力机制的SAR图像飞机检测", 《***工程与电子技术》, vol. 43, no. 11, 16 June 2021 (2021-06-16) * |
| 邵嘉琦 等: "基于CNN的不平衡SAR图像舰船目标识别", 《电光与控制》, vol. 26, no. 9, 31 December 2019 (2019-12-31) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114581766A (en) * | 2022-01-19 | 2022-06-03 | 南京航空航天大学 | Target classification method based on three-channel sparse SAR image data |
| CN115223398A (en) * | 2022-07-14 | 2022-10-21 | 集美大学 | Nuclear adaptive fractional order complex value neural network AIS interpolation method considering channel constraint |
| CN115223398B (en) * | 2022-07-14 | 2023-09-19 | 集美大学 | Nuclear self-adaptive fractional order complex value neural network AIS interpolation method considering channel constraint |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113537020B (en) | 2022-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860670B (en) | Domain adaptive model training method, image detection method, device, equipment and medium | |
| CN111860147B (en) | Pedestrian re-identification model optimization processing method and device and computer equipment | |
| US20200372243A1 (en) | Image processing method and apparatus, facial recognition method and apparatus, and computer device | |
| CN111080628A (en) | Image tampering detection method and device, computer equipment and storage medium | |
| CN108427927B (en) | Object re-recognition method and apparatus, electronic device, program, and storage medium | |
| CN113537020B (en) | Complex SAR image target identification method based on improved neural network | |
| CN108229673B (en) | Convolutional neural network processing method and device and electronic equipment | |
| CN111191533A (en) | Pedestrian re-identification processing method and device, computer equipment and storage medium | |
| CN113449586A (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN111047088A (en) | Prediction image acquisition method and device, computer equipment and storage medium | |
| CN114266897A (en) | Method and device for predicting pox types, electronic equipment and storage medium | |
| CN111860582B (en) | Image classification model construction method and device, computer equipment and storage medium | |
| CN112115937A (en) | Target identification method and device, computer equipment and storage medium | |
| CN110807463B (en) | Image segmentation method and device, computer equipment and storage medium | |
| CN110222752B (en) | Image processing method, system, computer device, storage medium and chip | |
| CN114821736A (en) | Multi-modal face recognition method, device, equipment and medium based on contrast learning | |
| CN111046755A (en) | Character recognition method, character recognition device, computer equipment and computer-readable storage medium | |
| CN111612732A (en) | Image quality evaluation method, image quality evaluation device, computer equipment and storage medium | |
| CN110889432A (en) | Feature point matching method and device, computer equipment and storage medium | |
| US20210374541A1 (en) | Information processing method and recording medium | |
| CN111462059B (en) | Parallel processing method and device for intelligent target detection of fetal ultrasonic image | |
| CN109934870B (en) | Target detection method, device, equipment, computer equipment and storage medium | |
| CN113256556A (en) | Image selection method and device | |
| CN113065617A (en) | Object recognition method, object recognition device, computer equipment and storage medium | |
| CN113762249A (en) | Image attack detection and image attack detection model training method and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |