Construction of radar image recognition model and radar image target recognition method

Through feature preprocessing and dictionary learning of radar image samples, sparse dictionary and neural network models are constructed, and the problems of radar image recognition accuracy and robustness are solved, and the precise recognition of high-resolution SAR images is achieved.

CN114463625BActive Publication Date: 2024-11-26CHINA TELECOM CLOUD TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111613160.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-27
Publication Date
2024-11-26
Estimated Expiration
2041-12-27

AI Technical Summary

Technical Problem

In the prior art, radar image recognition has low accuracy and robustness, making it difficult to effectively identify targets in high-resolution SAR images, especially due to complexity caused by enlarged imaging areas, coherent spot noise and azimuth changes.

Method used

By obtaining the original radar image samples, it is divided into target area and shadow area samples, dictionary learning, building sparse dictionary and neural network models, using sparse dictionary and models for image recognition, removing noise and background interference, and enhancing classification information .

Benefits of technology

It improves the accuracy and robustness of radar image recognition, reduces the influence of noise and background clutter, and makes image classification more refined and reliable.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114463625B_ABST
    Figure CN114463625B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention provides a radar image recognition model construction, radar image target recognition method, device and equipment, and the radar image recognition model construction method includes: obtaining original radar image samples, and performing feature preprocessing based on the original radar image samples to obtain radar original feature preprocessing image samples; performing image segmentation on the original radar feature preprocessing image samples to obtain target area image samples and shadow area image samples; performing dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary; and constructing a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary. By performing feature processing on the original radar image, the influence of irrelevant information in the SAR image on target classification is weakened, and important classification information is enhanced, so that the information classification contained in the original radar image is more refined.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of radar automatic recognition technology, and in particular to the construction of a radar image recognition model and a radar image target recognition method. Background Art

[0002] Compared with ordinary optical imaging systems, synthetic aperture radar (SAR) has a strong imaging advantage for ground targets, especially ground stationary targets. As an important means of target detection, it has received extensive attention and research. With the continuous innovation and development of SAR imaging systems, various SAR systems for different applications have begun to be enriched, and the imaging functions and effects have been continuously enhanced and improved. We have obtained a large amount of detection data using these powerful imaging systems. However, due to the application of high-resolution SAR, the expansion of imaging areas, and the diversity of imaging modes, traditional manual recognition methods have encountered challenges: on the one hand, the expansion of SAR imaging area has led to a sharp increase in the amount of data obtained by the acquisition system; on the other hand, SAR images have a large amount of coherent speckle noise, and the imaging results of the same target at different azimuths are very different, that is, synthetic aperture radar is sensitive to changes in target azimuth. The huge amount of data and complex image characteristics make manual interpretation of SAR image targets low in accuracy and robustness, and less and less practical, making it difficult to apply in practice. Summary of the invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the defects of low image recognition accuracy and robustness in the prior art, thereby providing a radar image recognition model construction and a radar image target recognition method.

[0004] According to a first aspect, an embodiment of the present invention provides a method for constructing a radar image recognition model, comprising: obtaining original radar image samples, and performing feature preprocessing based on the original radar image samples to obtain radar original feature preprocessed image samples; performing image segmentation on the original radar feature preprocessed image samples based on a preset segmentation threshold to obtain target area image samples and shadow area image samples; performing dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary; and constructing a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary.

[0005] Optionally, the dictionary learning based on the target area image samples and the shadow area image samples includes: performing image enhancement based on the target area image to obtain a target area enhanced image; performing centroid registration based on the target area enhanced image to obtain a target area centroid registered image; performing normalization processing based on the target area centroid registered image to obtain a target area normalized processed image; performing dictionary learning based on the target area normalized processed image and the shadow area image samples to obtain a sparse dictionary.

[0006] Optionally, performing feature preprocessing based on the original radar image to obtain the radar original feature preprocessed image includes: extracting original radar image data based on the original radar image samples; performing logarithmic transformation processing based on the original radar image data to obtain the radar original feature preprocessed image samples.

[0007] According to a second aspect, an embodiment of the present invention provides a radar image training device, including: an original image acquisition module, used to acquire original radar image samples, and perform feature preprocessing based on the original radar image samples to obtain radar original feature preprocessing image samples; an original image segmentation module, used to perform image segmentation on the original radar feature preprocessing image samples based on a preset segmentation threshold to obtain target area image samples and shadow area image samples; a dictionary learning module, used to perform dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary; a model construction module, used to construct a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary.

[0008] According to a third aspect, an embodiment of the present invention provides a radar image target recognition method, comprising: acquiring an original radar image, and performing feature preprocessing based on the original radar image to obtain a radar sample feature preprocessing image; acquiring target area feature information and shadow area feature information based on the radar sample feature preprocessing image; identifying the category of the original radar image based on the target area feature information and shadow area feature information and a preset radar image recognition model, wherein the preset radar image recognition model is constructed using the radar image recognition model construction method described in the first aspect or any one of the embodiments.

[0009] Optionally, the identifying the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model includes: obtaining sparse representation coefficients of the target area image samples and the shadow area image samples under the sparse dictionary based on a preset neural network model for image recognition and the sparse dictionary; reconstructing the target area information and the shadow area information of the original radar image samples based on the sparse coefficients, and judging the category of the original radar image samples according to the reconstruction error.

[0010] According to a fourth aspect, an embodiment of the present invention provides a radar image target recognition device, comprising: a sample acquisition module, used to acquire a radar sample image, and perform feature preprocessing based on the radar sample image to obtain a radar sample feature preprocessing image; a sample processing module, used to acquire target area feature information and shadow area feature information based on the radar sample feature preprocessing image; a model construction module, used to construct a joint sparse model based on the target area feature information and the shadow area feature information; an image recognition module, used to recognize the radar sample image based on the joint sparse model and a sparse dictionary.

[0011] According to a fifth aspect, a computer device comprises: a communication unit, a memory and a processor, wherein the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the steps of the method described in the first aspect, the second aspect or any optional embodiment by executing the computer instructions.

[0012] According to a sixth aspect, a computer-readable storage medium is provided, characterized in that the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the steps of the method described in the first aspect, the second aspect or any optional embodiment.

[0013] The technical solution of the present invention has the following advantages:

[0014] A method for constructing a radar image recognition model provided by an embodiment of the present invention comprises the following steps: obtaining original radar image samples, and performing feature preprocessing on the original radar image samples to obtain radar original feature preprocessing image samples, performing image segmentation on the original radar feature preprocessing image samples according to a preset segmentation threshold to obtain target area image samples and shadow area image samples, performing dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary, and using the sparse dictionary and a preset neural network model for image recognition to construct a radar image recognition model. The embodiment of the present invention removes noise, clutter and excessive background areas in the original radar image by processing the original radar prediction samples and performing dictionary learning based on the processed images, thereby weakening the influence of irrelevant information in the SAR image on target classification, enhancing important classification information, and making the information classification contained in the original radar image more refined.

[0015] A radar image target recognition method provided by an embodiment of the present invention comprises the following steps: obtaining an original radar image, and performing feature preprocessing on the original radar image to obtain a radar sample feature preprocessing image, extracting target area feature information and shadow area feature information contained in the image based on the radar sample feature preprocessing image, and performing recognition and classification of the radar image using a radar image recognition model and the target area feature information and shadow area feature information. The embodiment of the present invention can more accurately recognize and classify the original radar image by using the radar image recognition model to recognize and classify the original radar image. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 is a flowchart of a specific example of a method for constructing a radar image recognition model in an embodiment of the present invention;

[0018] Figure 2 is an example schematic diagram of a two-dimensional histogram in a method for constructing a radar image recognition model in an embodiment of the present invention;

[0019] Figure 3 is an example schematic diagram of a dynamic active set in a method for constructing a radar image recognition model in an embodiment of the present invention;

[0020] Figure 4is a principle block diagram of a device for constructing a radar image recognition model in an embodiment of the present invention;

[0021] Figure 5 A flowchart of a specific example of a radar image target recognition method according to an embodiment of the present invention;

[0022] Figure 6 It is a principle block diagram of a device for constructing a radar image target recognition model in an embodiment of the present invention;

[0023] Figure 7 The figure is a schematic diagram of an example structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0026] The embodiment of the present invention builds a radar image recognition model based on the original radar image, and uses the radar image recognition model to accurately identify and classify the original radar image.

[0027] Figure 1 A method for constructing a radar image recognition model according to an embodiment of the present invention is shown, comprising the following steps:

[0028] S100: Acquire original radar image samples, and perform feature preprocessing based on the original radar image samples to obtain radar original feature preprocessed image samples.

[0029] Specifically, an original radar image sample is obtained, wherein the original radar image sample contains coherent speckle noise and background clutter, and the original radar image target scattering center is uneven, so feature preprocessing is performed based on the original radar image sample to eliminate the coherent speckle noise and background clutter, and obtain the radar original feature preprocessed image sample.

[0030] S200: performing image segmentation on the original radar feature preprocessed image samples to obtain target area image samples and shadow area image samples.

[0031] Specifically, a two-dimensional histogram of the mean grayscale of image points and the mean grayscale of regions is obtained based on the original radar feature preprocessed image samples, a segmentation threshold is set based on the two-dimensional histogram, and the original radar feature preprocessed image samples are segmented based on the segmentation threshold to obtain target region image samples and shadow region image samples. The two-dimensional histogram is as follows: Figure 2 shown.

[0032] In practical applications, the segmentation threshold can be improved by, for example, a genetic algorithm to improve the efficiency of obtaining the segmentation threshold based on a two-dimensional histogram. Specifically, based on the two-dimensional histogram, the gray level L of the radar original feature preprocessing image sample is set to 256, and the threshold is obtained using 8-bit binary coding, so (s, t) is encoded as a 16-bit gene string α={α0,…,α 7, ,α8,…,α 15}, where the high 8 bits represent s and the low 8 bits represent t; initialize the gene string to obtain an initialization population, where the initialization population can be set to 30 to 50, for example; based on the decoding of the initialization population, calculate the fitness of each gene string in the initialization population, and the formula is g=0,1,..,G, where g is the current evolutionary generation and G is the maximum evolutionary generation; if the solution calculated by the formula converges sufficiently or reaches the evolutionary generation, the threshold represented by the code of the individual with the highest fitness is used as the segmentation threshold, and the radar original feature preprocessed image sample is segmented; if the evolutionary generation is not reached or the solution is divergent, selection, crossover, and mutation operations are performed based on the gene string to generate a new population, and then the segmentation threshold is calculated again. The selection operation is the most commonly used roulette in genetic algorithms. First, according to the elite strategy, the top 10% of the excellent individuals in fitness are directly copied to the next generation, and then the remaining individuals are selected based on the roulette method. The higher the individual fitness, the greater the chance of being selected and the more likely it is to be inherited to the next generation; the crossover operation adopts a double-point crossover, and a crossover point is randomly generated based on the high 8 bits and the low 8 bits of the coding string. Generally, the crossover probability P c is a given constant. In this embodiment, P c Set it to 0.8 to obtain more individuals, and then set P c is set to 0.6 to establish the best model when the search range is reduced; the mutation operation is to control the proportion of new genes introduced into the new population, and the mutation rate P m For example, the value can be 0.01 to 0.2. g=0,1,..,G,where P m_min =P b ,P m_max =10P m_min , basic mutation rate P b is defined as: Where populations is the number of populations, which is set to 30 in the embodiment of the present invention, and bits is the number of gene strings, which is set to 16 in the embodiment of the present invention, so as to obtain P b ≈0.015, setting the maximum evolutionary number G to 100, we can get P m =0.15-0.000054(g-50) 2 .

[0033] S300: Perform dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary.

[0034] Specifically, label information is extracted based on the target area image samples, and a target area dictionary and a shadow area dictionary with "reconstruction" and "discrimination" are learned according to the label information, and the target area dictionary and the shadow area dictionary constitute a sparse dictionary.

[0035] S400: Constructing a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary.

[0036] Specifically, the neural network model for image recognition is a joint sparse representation model (JSR), which combines the sparse representation problem of the original radar feature preprocessed image, that is, st‖x k ‖0≤S, in, is the original radar image sample Y = [y1, y2, ..., y K ] is a sparse coefficient matrix, A k is the dictionary of the kth feature, S is the sparsity, K is the number of features, K=2 is a joint sparse representation model, and a radar image recognition model is constructed based on the joint sparse representation model and the sparse dictionary.

[0037] In an embodiment of the present invention, an original radar image sample is obtained, and feature preprocessing is performed based on the original radar image sample to eliminate coherent speckle noise and background clutter to obtain a radar original feature preprocessed image sample, and the radar original feature preprocessed image is segmented, and dictionary learning is performed based on the segmented target area image samples and shadow area image samples to obtain a sparse dictionary, and a radar image recognition model is constructed using the sparse dictionary and a preset neural network model for image recognition. In this embodiment, the original radar image samples are processed and dictionary learning is performed, so as to overcome the interference of noise and background clutter in the original radar image samples, convert the original radar image samples into samples that are conducive to recognition, and thereby construct a radar image recognition model.

[0038] In an optional embodiment of the present invention, the dictionary learning based on the target area image samples and the shadow area image samples in step S300 includes the following steps:

[0039] (1) Performing image enhancement based on the target area image to obtain a target area enhanced image.

[0040] Specifically, the target area image is segmented again to obtain a binary image B, mask processing is performed based on the binary image and the original radar image sample to obtain a binary mask image M, and the target area image is enhanced based on the binary mask image and power transformation, and the method is H(x, y) = [M(x, y)] α , where M and H refer to the images before and after the transformation, respectively, and α is a constant greater than 0, so as to obtain the enhanced image of the target area.

[0041] (2) Performing centroid registration based on the target area enhanced image to obtain a target area centroid registration image.

[0042] Specifically, the centroid registration is continued based on the target area enhanced image, and the centroid registration is performed by Get, where h pq is the p+q-order moment of image H, which is defined as h pq =∑ x ∑ y x p y p H(x,y). In practical applications, the target area in the original radar image is small, and the background area is large. Therefore, for example, a square with a side length of 50 pixels in the original radar image can be set as the important area. The original radar image is centered at the centroid of the target area image and expands to a 50×50 square in all directions.

[0043] (3) Performing normalization processing on the target area centroid registration image to obtain a target area normalized processing image.

[0044] Specifically, since the radar image has different ranges and different echo intensities during the acquisition process, the overall brightness of the original radar image acquired is different. Therefore, it is necessary to normalize the segmented target area image. The normalization is performed based on the target area centroid registration image. The normalization makes the target area image meet ∑ x ∑ y |H(x,y)| 2 =1, the normalization formula is The normalized processed image of the target area is obtained through the above formula.

[0045] (4) Perform dictionary learning based on the normalized image of the target area and the image samples of the shadow area to obtain a sparse dictionary.

[0046] Specifically, centroid registration is performed based on the shadow area image, with the centroid of the shadow area image as the center, the centroid registration formula of the target area is used to perform the shadow area centroid registration to obtain a shadow area centroid registration image, and dictionary learning is performed based on the shadow area centroid registration image and the normalized image of the target area to obtain a sparse dictionary.

[0047] In the embodiment of the present invention, the target area image and the shadow area image are subjected to image processing again to remove the noise, clutter and excessive background areas contained in the target area image and the shadow area image, and at the same time, the influence of irrelevant information in the original radar image on target recognition and classification is weakened, and important classification information is enhanced, and the target position of the original radar image is registered, so that the target recognition and classification are more refined.

[0048] In an optional embodiment of the present invention, the step S100 described above performs feature preprocessing based on the original radar image to obtain the radar original feature preprocessed image, including the following steps:

[0049] (1) extracting original radar image data based on the original radar image samples;

[0050] (2) Performing logarithmic transformation processing on the original radar image data to obtain the radar original feature preprocessing image sample.

[0051] In an embodiment of the present invention, the original radar image data is extracted using the acquired original radar image samples, and the original radar image data is subjected to logarithmic transformation processing, and the logarithmic transformation formula is K(x,y)=10log[G(x,y)+Y]+X, where G represents the amplitude matrix of the original radar image sample, so as to obtain the radar original feature preprocessing image sample. In an embodiment of the present invention, the acquired original radar image samples are subjected to logarithmic transformation processing, so that the speckle noise of the original radar image can be suppressed, so that the acquired image histogram is more similar to a Gaussian distribution. In practical applications, since the logarithm is undefined at 0, an arbitrary constant is added to the original radar image, and Y in the above formula can be, for example, 0.001. In order to ensure that the pixel value is non-negative, a corresponding constant is added, and X in the above formula can be, for example, 30.

[0052] In an optional embodiment of the present invention, the step S400 described in the above-mentioned step of constructing a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary includes the following steps:

[0053] (1) Training is performed based on the sparse dictionary to obtain a label consistent item model.

[0054] Specifically, the sparse dictionary is learned using a supervised learning algorithm. The sparse dictionary includes a target area dictionary and a shadow area dictionary. The corresponding label information is extracted based on the target area dictionary and the shadow area dictionary, and combined with the reconstruction error and the classification error to obtain the label consistency item models LC-KSVD1 and LC-KSVD2. The expression of LC-KSVD1 is: where α controls the relative contribution between the reconstruction error and label consistency regularization, D∈R M×M is a linear transformation matrix, X is the sparse matrix of the original radar image, represents the discriminative sparse coding error, Make the transformed sparse code DX close to the discriminative sparse code Q, so that the sparse coding obtained by learning the target area dictionary and the shadow area dictionary can promote the consistency of the extracted labels, Q = [q1, q2, ..., q N ]∈R M×N Used to control the sparsity mode, With feature sample f i Related: When f i and the mth dictionary atom a m When they are of the same type, otherwise, The expression of LC-KSVD2 is where α and β are scalars that control the relative contributions of the corresponding terms, represents the classification error, W represents the classifier parameters, H = [h1,h2,…,h N ]∈R P×N is the category label of the input signal F, P is the total number of categories, and h i =[0,..,0,1,0,…,0] T ∈R P is the training sample f i The label vector of the non-zero entry is f i Category.

[0055] For example, assuming A = [a1, a2, a3, a4, a5, a6], F = [f1, f2, f3, f4], where f1, a1, a2 are the first category; f2, a3, a4 are the second category, and f3, f4, a5, a6 are the third category, then Q can be defined as

[0056] Exemplarily, the basic process of constructing LC-KSVD2 is shown in Table 1 below:

[0057] Table 1

[0058]

[0059] The optimal solution for all parameters in the LC-KSVD2 is obtained by using the K-SVD algorithm, namely: Where Q is a constant matrix representing the sparse dictionary sub-labels. In practical applications, for example, but Thus, the solution of the label consistency model is obtained using the traditional K-SVD algorithm.

[0060] For example, the LC-KSVD calculation requires initialization parameters A (0) , D (0) and W (0) For A (0) , several K-SVD iterations are performed in each class, and then the outputs of each K-SVD are combined together. Each item a in the dictionary m The class labels of are initialized based on the relevant classes and remain unchanged throughout the sparse dictionary training process, and do not change with a m In practical applications, a multivariate ridge regression model, quadratic loss and l2 norm regularization are used to initialize D (0) , the calculation formula is: And by calculation, we can get D = QX T (XX T +λ1I) -1 ; For W (0) , using the above multivariate ridge regression model formula, we can get the following solution: W = HX T (XX T +λ2I) -1 , preset initialization A (0) , use the K-SVD algorithm to calculate the sparse coding coefficient X of the training signal F, and then use X to initialize the formula D = QX T (XX T +λ1I) -1 D (0) and W = HX T (XX T +λ2I) -1 W (0) .

[0061] (2) Constructing a radar image recognition model based on a preset neural network model for image recognition and the label consistency item model.

[0062] Specifically, the preset neural network model for image recognition is a joint sparse representation model (JSR), which combines the sparse representation problem in radar images, namely st‖x k ‖0≤S, in is the test sample Y=[y1,y2,…,y K ] is a sparse coefficient matrix, A k is the dictionary of the kth feature, S is the sparsity, K is the number of features, and K=2 is the joint sparse representation model. A radar image recognition model is constructed based on the joint sparse representation model and the label consistency model.

[0063] Exemplarily, the relationship between different features in the original radar image is represented based on the dynamic active set in the joint sparse representation model, g s ∈R K ,s=1,2,…,S, in the sparse matrix X, each dynamic active set refers to the row index of a group of coefficients belonging to the same class, as described Figure 3 In practical applications, each collected original radar image is used to extract the target area features and shadow area features, that is, K = 2, where X = [X1, X2] represents the coefficient matrix, X k (k=1,2) represents the kth feature y of the test sample Y k The corresponding sparse representation coefficient vector, the blank squares indicate that the sparse coefficient value is zero, and the numbers in the non-blank squares are the coefficient values. Figure 3 In the above, g1, g2 and g3 are three dynamically active sets. According to the definition of dynamically active sets, we know that g1 = (1, 2), g2 = (5, 5), g2 = (8, 7). The above dynamically active sets are jointly activated, and the radar image recognition model to be optimized is constructed as where x gs =X(g s )=[X(g s (1),1),X(g s (2),2),…,X(g s (K),K)] T ∈R K , where x gs refers to the sth dynamically active set g s The vector formed by the set of related coefficients. g s (k) represents the row index of the training sample selected in the kth column of the coefficient matrix A for the sth dynamic active set. Figure 3 For example, Figure 3 The coefficient vector corresponding to g1 in The coefficient vector corresponding to g2 The coefficient vector corresponding to g3 Therefore, the radar image recognition model to be optimized can be simplified as follows: Where S is the sparsity, which means that at each x k The number of non-zero elements in .

[0064] Exemplarily, the radar image recognition model to be optimized can be optimized in combination with the extended sparse representation method, and its optimization objective function can be expressed as The optimized model is a radar image recognition model

[0065] Exemplarily, in solving the radar image recognition model in the embodiment of the present invention, the reconstruction error R=Y and the dynamic active set are first initialized. Then, after multiple iterations of optimization, the solution is solved until the stopping condition is met. The solution process includes: (i) selecting a new candidate set based on the current error; (ii) merging the newly selected dynamic active set with the previous dynamic active set; (ii i) using the dynamic active set obtained in the previous step to solve the corresponding sparse coefficient; (iv) based on the latest estimated sparse representation coefficient, reducing the sparsity of the mixed set to a specific sparse level; (v) updating the error; (vi) judging whether the current iteration stopping condition is met, if so, stop, if not, go to (i).

[0066] Exemplarily, at each iteration of the radar image recognition model, i.e., the above steps (i) and (iv), the coefficient matrix Z∈R N×K , select L candidate dynamic active sets from them, and use the candidate dynamic active sets to construct a matrix similar to Z And solve

[0067] For example, the joint dynamic sparse matching (JDSmapping) algorithm is used to solve and obtain I L =P JDS (Z,L), where the index matrix I L ∈R L×K The top-L dynamic active sets containing all K features update X and E in each iteration, which is the process of step (iii): k ′=(A k T A k ) -1 A k T (y k -e k )、e k ′=y k -Ak x k .

[0068] In an embodiment of the present invention, the sparse dictionary is trained to extract label information from the original radar image, and a label consistent item model is constructed based on the label information. A radar image recognition model to be optimized is constructed based on the consistent item label model and a preset joint sparse representation model, and the radar image recognition model to be optimized is optimized to obtain a radar image recognition model, so that the feature information contained in the original radar image samples obtained by dictionary learning is processed into easily recognizable feature information, thereby reducing image recognition errors and solving problems such as noise, occlusion, and damage in the original radar image.

[0069] like Figure 4 As shown, an embodiment of the present invention provides a device for constructing a radar image recognition model, including: an original image acquisition module 1, an original image segmentation module 2, a dictionary learning module 3, and a model construction module 4, wherein:

[0070] The original image acquisition module 1 is used to acquire original radar image samples, and perform feature preprocessing based on the original radar image samples to obtain radar original feature preprocessed image samples. For details, please refer to the relevant description of step S100 of any of the above method embodiments;

[0071] The original image segmentation module 2 is used to perform image segmentation on the original radar feature preprocessed image sample based on a preset segmentation threshold to obtain a target area image sample and a shadow area image sample. For details, please refer to the relevant description of step S200 of any of the above method embodiments;

[0072] A dictionary learning module 3 is used to perform dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary. For details, please refer to the relevant description of step S300 in any of the above method embodiments;

[0073] The model building module 4 is used to build a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary. For details, please refer to the relevant description of step S400 of any of the above method embodiments.

[0074] In an embodiment of the present invention, an original radar image sample is obtained, and feature preprocessing is performed based on the original radar image sample to eliminate coherent speckle noise and background clutter to obtain a radar original feature preprocessed image sample, and the radar original feature preprocessed image is segmented, and dictionary learning is performed based on the segmented target area image samples and shadow area image samples to obtain a sparse dictionary, and a radar image recognition model is constructed using the sparse dictionary and a preset neural network model for image recognition. In this embodiment, the original radar image samples are processed and dictionary learning is performed, so as to overcome the interference of noise and background clutter in the original radar image samples, convert the original radar image samples into samples that are conducive to recognition, and thereby construct a radar image recognition model.

[0075] like Figure 5 As shown, an embodiment of the present invention provides a radar image target recognition method, the method comprising the following steps:

[0076] S500: Acquire an original radar image, and perform feature preprocessing based on the original radar image to obtain a radar sample feature preprocessing image.

[0077] Specifically, an original radar image is obtained, the original radar image contains coherent speckle noise and background clutter, and the target scattering center of the original radar image is uneven, so feature preprocessing is performed based on the original radar image to eliminate the coherent speckle noise and background clutter, and obtain a radar sample feature preprocessing image.

[0078] S600: Acquire target area feature information and shadow area feature information based on the radar sample feature preprocessing image.

[0079] Specifically, a two-dimensional histogram of image point grayscale and regional grayscale mean is obtained based on the radar sample feature preprocessed image, a segmentation threshold is set based on the two-dimensional histogram, and the radar sample feature preprocessed image is segmented based on the segmentation threshold to obtain target area feature information and shadow area feature information.

[0080] S700: Identify the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model, wherein the preset radar image recognition model is generated using the radar image recognition model construction method described in the above embodiment.

[0081] Specifically, the radar sample is reconstructed based on the target area feature information and the shadow area feature information and a preset radar recognition model, and the category of the radar sample is determined according to the reconstruction error. The preset radar recognition model is generated by training using the facial expression analysis method described in the above embodiments. For details, please refer to the relevant description of any of the above method embodiments.

[0082] In an embodiment of the present invention, the collected radar image samples are reconstructed by using a preset radar recognition model to determine the category of the radar sample, which can reduce the impact of noise, background clutter, etc. contained in the initial state of the radar sample and improve the accuracy of determining the category of the radar image.

[0083] In an optional embodiment of the present invention, the step S700 of identifying the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model includes the following steps:

[0084] (1) based on a preset neural network model for image recognition and the sparse dictionary, obtaining sparse representation coefficients of the target area image samples and the shadow area image samples under the sparse dictionary;

[0085] (2) reconstructing the target area information and the shadow area information of the original radar image sample based on the sparse coefficient, and determining the category of the original radar image sample according to the reconstruction error.

[0086] Exemplarily, in an embodiment of the present invention, the original radar sample is set to Y=[y1, y2], and the preset neural network model for image recognition is used to represent the target area image sample and the shadow area image sample of the original radar sample data, and the sparse coefficient thereof is calculated. The formula used is: and coefficient correction vector Based on the obtained sparse coefficients, the target area feature information and the shadow area feature information in the original radar image are reconstructed. The calculation formula for the reconstruction is: in is the category label of the radar sample, w k is the reconstruction parameter of the kth feature, and the error weight is set according to the importance of the feature. The operator δ c (·) means saving the rows corresponding to class c and setting the other rows to zero.

[0087] In the embodiment of the present invention, the target area and the shadow area in the radar sample are reconstructed, and the original radar image is identified and classified according to the reconstruction result, so that the original radar image can be accurately identified and the error can be reduced.

[0088] like Figure 6 As shown, the embodiment of the present invention further provides a radar image target recognition device, which is characterized by comprising: a sample acquisition module 5, a sample processing module 6, and an image recognition module 7, wherein:

[0089] The sample acquisition module 5 is used to acquire an original radar image, and perform feature preprocessing based on the original radar image to obtain a radar sample feature preprocessing image. For details, please refer to the relevant description of step S500 in any of the above method embodiments;

[0090] A sample processing module 6, used for obtaining target area feature information and shadow area feature information based on the radar sample feature preprocessing image, for details, please refer to the relevant description of step S600 of any of the above method embodiments;

[0091] The image recognition module 7 is used to identify the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model. The preset radar image recognition model is generated using the radar image recognition model construction method described in the above embodiment. For details, please refer to the relevant description of step S700 of any of the above method embodiments.

[0092] In an embodiment of the present invention, the collected radar image samples are reconstructed by using a preset radar recognition model to determine the category of the radar sample, which can reduce the impact of noise, background clutter, etc. contained in the initial state of the radar sample and improve the accuracy of determining the category of the radar image.

[0093] The specific limitations and beneficial effects of the radar image recognition model construction device and the radar image target recognition device can be found in the above-mentioned limitations on the radar image recognition model construction method and the radar image target recognition method, which will not be repeated here. The various modules of the above-mentioned radar image recognition model construction device and the radar image target recognition device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0094] The embodiment of the present invention also provides a computer device, such as Figure 7 As shown, Figure 7It is a structural diagram of a computer device provided by an optional embodiment of the present invention, and the computer device may include at least one processor 41, at least one communication interface 42, at least one communication bus 43 and at least one memory 44, wherein the communication interface 42 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 42 may also include a standard wired interface and a wireless interface. The memory 44 may be a high-speed RAM memory (Random Access Memory, volatile random access memory) or a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 44 may optionally be at least one storage device located away from the aforementioned processor 41. The processor 41 may be combined with Figure 4 , Figure 6 In the described device, the memory 44 stores an application program, and the processor 41 calls the program code stored in the memory 44 to execute the steps of the radar image recognition model construction method and the radar image target recognition method of any of the above method embodiments.

[0095] The communication bus 43 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus 43 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0096] Among them, the memory 44 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 44 may also include a combination of the above types of memory.

[0097] The processor 41 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and a NP.

[0098] The processor 41 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0099] Optionally, the memory 44 is also used to store program instructions. The processor 41 can call the program instructions to implement the method described in any embodiment of the present invention.

[0100] The embodiment of the present invention further provides a non-transitory computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the method described in any of the above method embodiments. The storage medium may be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above types of memory.

[0101] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for constructing a radar image recognition model, characterized in that: The steps include: Acquire original radar image samples, and perform feature preprocessing based on the original radar image samples to obtain original radar feature preprocessed image samples; Performing image segmentation on the original radar feature preprocessed image samples to obtain target area image samples and shadow area image samples; Perform dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary; Building a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary; The performing dictionary learning based on the target area image samples and the shadow area image samples includes: Performing image enhancement based on the target area image to obtain a target area enhanced image; Performing centroid registration based on the target area enhanced image to obtain a target area centroid registered image; Performing normalization processing based on the target area centroid registration image to obtain a target area normalized processed image; Performing dictionary learning based on the normalized image of the target area and the image samples of the shadow area to obtain a sparse dictionary; The radar image recognition model is constructed based on the preset neural network model for image recognition and the sparse dictionary, including: Performing training based on the sparse dictionary to obtain a label consistent item model; A radar image recognition model is constructed based on a preset neural network model for image recognition and the label consistency item model.

2. The method for constructing a radar image recognition model according to claim 1, characterized in that: The performing feature preprocessing based on the original radar image to obtain the radar original feature preprocessed image includes: extracting raw radar image data based on the raw radar image samples; Logarithmic transformation is performed based on the original radar image data to obtain the original radar feature preprocessed image samples.

3. A device for constructing a radar image recognition model, characterized in that: include: The original image acquisition module is used to acquire original radar image samples and perform feature preprocessing based on the original radar image samples to obtain original radar feature preprocessing image samples; An original image segmentation module is used to perform image segmentation on the original radar feature preprocessed image sample based on a preset segmentation threshold to obtain a target area image sample and a shadow area image sample; A dictionary learning module, used for performing dictionary learning based on the target area image samples and the shadow area image samples to obtain a sparse dictionary; A model building module, used to build a radar image recognition model based on a preset neural network model for image recognition and the sparse dictionary; The dictionary learning module is also used for: Performing image enhancement based on the target area image to obtain a target area enhanced image; Performing centroid registration based on the target area enhanced image to obtain a target area centroid registered image; Performing normalization processing based on the target area centroid registration image to obtain a target area normalized processed image; Performing dictionary learning based on the normalized image of the target area and the image samples of the shadow area to obtain a sparse dictionary; The model building module is also used to: Performing training based on the sparse dictionary to obtain a label consistent item model; A radar image recognition model is constructed based on a preset neural network model for image recognition and the label consistency item model.

4. A radar image target recognition method, characterized in that: include: Acquire an original radar image, and perform feature preprocessing based on the original radar image to obtain a radar sample feature preprocessing image; Acquire target area feature information and shadow area feature information based on the radar sample feature preprocessing image; The category of the original radar image is identified based on the target area feature information and the shadow area feature information and a preset radar image recognition model, wherein the preset radar image recognition model is constructed using the radar image recognition model construction method as described in claim 1 or 2.

5. The radar image target recognition method according to claim 4, characterized in that: The identifying the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model includes: Based on a preset neural network model for image recognition and the sparse dictionary, sparse representation coefficients of the target area image samples and the shadow area image samples under the sparse dictionary are obtained; The target area information and the shadow area information of the original radar image sample are reconstructed based on the sparse representation coefficient, and the category of the original radar image sample is determined according to the reconstruction error.

6. A radar image target recognition device, characterized in that: include: A sample acquisition module, used to acquire an original radar image, and perform feature preprocessing based on the original radar image to obtain a radar sample feature preprocessing image; A sample processing module, used for obtaining target area feature information and shadow area feature information based on the radar sample feature preprocessing image; An image recognition module is used to identify the category of the original radar image based on the target area feature information and the shadow area feature information and a preset radar image recognition model, wherein the preset radar image recognition model is constructed using the radar image recognition model construction method as described in claim 1 or 2.

7. A computer device, characterized in that: include: A communication unit, a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the steps of the method described in claim 1 or 2 and the steps of the method described in claim 4 or 5 by executing the computer instructions.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute the steps of the method according to claim 1 or 2 and the steps of the method according to claim 4 or 5.