SAR (Synthetic Aperture Radar) feature analysis method and device for intelligent network hidden layer mapping

Abstract

The invention discloses an SAR (Synthetic Aperture Radar) feature analysis method and device for intelligent network hidden layer mapping, and belongs to the field of SAR feature extraction and recognition. The method comprises the following steps: inputting acquired SAR image data into a preset feature extraction network, and outputting hidden layer feature information and traditional feature information of the SAR image data; performing comparative analysis on the hidden layer feature information and the traditional feature information, and determining a contrast relationship between the hidden layer feature information and the traditional feature information; and determining key hidden layer features of the hidden layer feature information in the analysis result according to a preset requirement, and performing reverse activation mapping on the key hidden layer features to obtain intelligent identification distribution characteristics of the SAR image data. According to the method, the feature learning and interpretation capability of the deep learning network on the SAR image can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of SAR feature extraction and recognition, and in particular to a SAR feature analysis method and device oriented to intelligent network hidden layer mapping. Background Art

[0002] Synthetic Aperture Radar (SAR) is an active microwave remote sensing technology that achieves high-resolution imaging by transmitting electromagnetic waves and receiving reflected signals from targets, combined with the mobility and signal processing of mobile platforms (such as satellites, aircraft, and drones). Its core advantages lie in its all-weather and all-day observation capabilities and its highly sensitive detection of subtle surface changes, making it widely used in various fields.

[0003] In related technologies, due to the low efficiency of manual analysis of SAR image data and the difficulty in mapping the hidden features of the data with the electromagnetic scattering features during the deep learning process, the deep learning network's ability to learn and interpret the features of SAR images cannot meet application needs.

[0004] Based on this, there is an urgent need for a SAR feature analysis method and device for intelligent network hidden layer mapping to solve the above technical problems. Summary of the Invention

[0005] This invention provides a SAR feature analysis method and device for intelligent network hidden layer mapping, which can effectively improve the feature learning and interpretation capabilities of deep learning networks for SAR images. The technical solution is as follows:

[0006] In one aspect, a SAR feature analysis method for intelligent network hidden layer mapping is provided, the method comprising:

[0007] Inputting the acquired SAR image data into a preset feature extraction network, and outputting hidden layer feature information and traditional feature information of the SAR image data;

[0008] performing comparative analysis on the latent layer feature information and the traditional feature information to determine a comparison relationship between the latent layer feature information and the traditional feature information;

[0009] The key hidden layer features of the hidden layer feature information in the analysis result are determined according to preset requirements, and the key hidden layer features are reversely activated mapped to obtain the intelligent recognition distribution characteristics of the SAR image data.

[0010] On the other hand, a SAR feature analysis device for intelligent network hidden layer mapping is provided, the device comprising:

[0011] An extraction module is used to input the acquired SAR image data into a preset feature extraction network, and output hidden layer feature information and traditional feature information of the SAR image data;

[0012] An analysis module, configured to compare and analyze the latent layer feature information and the traditional feature information to determine a correspondence relationship between the latent layer feature information and the traditional feature information;

[0013] The mapping module is used to determine the key hidden layer features of the hidden layer feature information in the analysis result according to preset requirements, and perform inverse activation mapping on the key hidden layer features to obtain the intelligent recognition distribution characteristics of the SAR image data.

[0014] On the other hand, a computer device is provided, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the above-mentioned SAR feature analysis method for intelligent network hidden layer mapping.

[0015] On the other hand, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the SAR feature analysis method for intelligent network hidden layer mapping are implemented.

[0016] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned SAR feature analysis method for intelligent network hidden layer mapping.

[0017] The technical solution provided by the present invention can achieve at least the following beneficial effects: First, an intelligent recognition network based on SAR image data is constructed, and the network's hidden layer features are extracted after the data is input. Simultaneously, traditional feature information of the SAR target, including geometric features, texture features, and scattering center features, is effectively extracted for use in latent feature interpretation and analysis. Then, the hidden layer feature information extracted from the intelligent recognition network is compared and analyzed with traditional SAR features to select key hidden layer features. Finally, the key hidden layer feature maps are subjected to inverse activation mapping to obtain a complete analysis result of the SAR data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a SAR feature analysis method for intelligent network hidden layer mapping provided by one embodiment of the present invention;

[0020] Figure 2 is a feature extraction network structure diagram provided by an embodiment of the present invention;

[0021] Figure 3 1 is a schematic diagram of the SAR image geometric feature extraction results provided by one embodiment of the present invention;

[0022] Figure 4 2 is a schematic diagram of the GLCM description principle provided by an embodiment of the present invention;

[0023] Figure 5 1 is a schematic diagram of a SAR image texture feature extraction result provided by an embodiment of the present invention;

[0024] Figure 6 This is a schematic diagram of the SAR image scattering center extraction results provided by one embodiment of the present invention;

[0025] Figure 7 This is a schematic diagram of visualizing shallow network features provided by an embodiment of the present invention;

[0026] Figure 8 This is a schematic diagram of visualizing shallow features in a network provided by an embodiment of the present invention;

[0027] Figure 9 This is a schematic diagram of visualizing deep features in a network provided by an embodiment of the present invention;

[0028] Figure 10 This is a schematic diagram of visualizing deep features of a network provided by an embodiment of the present invention;

[0029] Figure 11 1 is a schematic diagram of the analysis of hidden layer feature mapping results provided by an embodiment of the present invention;

[0030] Figure 12 This is a structural diagram of a SAR feature analysis device for intelligent network hidden layer mapping provided by one embodiment of the present invention;

[0031] Figure 13 This is a hardware architecture diagram of a computer device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] As mentioned above, the non-intuitive characteristics of SAR image data lead to low efficiency and high false detection rate in manual interpretation.

[0034] Based on this, the present invention constructs an intelligent recognition network and determines the association between the hidden features and traditional features of SAR images, thereby improving the feature learning and interpretation capabilities of the deep learning network for SAR images.

[0035] The specific implementation of the above concept is described below.

[0036] Please refer to Figure 1 The embodiment of the present invention provides a SAR feature analysis method for intelligent network hidden layer mapping, the method comprising:

[0037] Step 100: input the acquired SAR image data into a preset feature extraction network, and output the latent feature information and traditional feature information of the SAR image data;

[0038] Step 102: performing comparative analysis on the latent layer feature information and the traditional feature information to determine a comparison relationship between the latent layer feature information and the traditional feature information;

[0039] Step 104 : determining key hidden layer features of the hidden layer feature information in the analysis result according to preset requirements, and performing inverse activation mapping on the key hidden layer features to obtain intelligent recognition distribution characteristics of the SAR image data.

[0040] Described below Figure 1 How to perform the steps shown.

[0041] First, with respect to step 100 , the acquired SAR image data is input into a preset feature extraction network, and the hidden layer feature information and traditional feature information of the SAR image data are output.

[0042] In the embodiment of the present invention, the feature extraction network is constructed based on the CNN network. Figure 2As shown in Figure 2, in a CNN, convolutional and pooling layers can be viewed as mappings from datasets to feature maps, while fully connected layers can be viewed as mappings from feature maps to sample labels. The connected layers are trained using the feature maps from the previous layers to generate a loss function. The error from this loss function is then backpropagated to the previous layers to adjust and update the network parameters.

[0043] In this embodiment of the present invention, the traditional feature information extracted includes the geometric features, texture features, and scattering center features of the SAR target. The geometric features are extracted by performing a coarse segmentation process on the filtered original image based on constant false alarm rate (CFAR); performing morphological filtering on the segmentation result based on area and distance to remove areas that do not meet preset requirements; and performing pixel-wise multiplication on the filtered result image and the original image to obtain the geometric features of the SAR image.

[0044] Specifically, for a given SAR image target, shape is one of its most important features. Since SAR images have the characteristic of point spread function energy accumulation in the electromagnetic scattering mechanism, using a mask method to describe the shape is more in line with the basic imaging principle of SAR images.

[0045] First, the original image is filtered and the constant false alarm rate (CFAR) method is used to perform rough target segmentation on the SAR / ISAR image. Assuming that the clutter obeys a negative exponential distribution, the detection threshold is

[0046] T=-μln(P fa )

[0047] Among them, P fa is the false alarm rate, μ is the mean value. For each point (i, j) in the SAR / ISAR image, if its pixel value is greater than the threshold T, it is judged as a target, otherwise it is considered as background.

[0048] Morphological filtering is performed on the segmentation results to remove non-target areas, reduce noise, smooth boundaries, remove small holes, etc.

[0049] Area-based filtering. First, remove isolated points from the constant false alarm detection result image, and then filter out points with an area smaller than the preset threshold T. A area.

[0050] Distance-based filtering. For the result image of the previous step, first find the area with the largest area and its centroid, then calculate the distance between each area and the area with the largest area, and finally filter out the areas with a distance greater than the preset threshold T from the area with the largest area. D area.

[0051] Mask. In order to obtain the intensity information of the target, the result image (binary image) obtained after filtering is multiplied by the original image pixel by pixel to obtain the final target intensity image, such as Figure 3 shown.

[0052] Texture descriptions of SAR images are multidimensional. Currently, the most commonly used texture descriptor is the Gray-Level Co-occurrence Matrix (GLCM). A GLCM describes the grayscale relationship between a pixel in a local or global region of an image and its adjacent pixels or pixels within a certain distance. Its depth reflects the global frequency of pixel pairs of a certain shape within the image.

[0053] like Figure 4 As shown in Figure 2, GLCM includes the following basic concepts:

[0054] ① Matrix size: If the grayscale level of the original image is not compressed, the size of the GLCM is the original grayscale level ^ 2; in practical applications, considering the computational efficiency of texture features and the storage of the GLCM matrix, the grayscale level of the original image is usually compressed first. For example, an 8-bit image with a grayscale level of 0-255 is compressed into a 5-bit image with a grayscale level of 0-31. The dimension of the corresponding co-occurrence matrix is reduced from 256*256 to 32*32.

[0055] ② Reference window: A window centered on the current pixel, usually with an odd size (3*3, 5*5, 7*7, etc.).

[0056] ③ Sliding Window: A window that moves with the reference window as the reference window, in the specified direction and step size. The size is the same as the reference window.

[0057] ④ Moving direction: The relative direction between the reference window and the moving window. The moving direction can be set arbitrarily, usually 0°, 45°, 90°, 135°

[0058] ⑤ Moving step: the pixel distance between the center pixel of the reference window and the center pixel of the sliding window

[0059] The GLCM texture feature extraction results of SAR container ships are as follows: Figure 5 shown.

[0060] The scattering center features are extracted in the following manner: Stolt interpolation processing is performed on the angle-frequency two-dimensional original echo of the acquired SAR image data, and the processed data is estimated to obtain a covariance matrix; the covariance matrix is eigendecomposed, and the decomposition result is estimated using the Gauss circle method to obtain the number of scattering centers; a preset two-dimensional multiple signal classification spectrum is searched and discretely searched respectively to obtain position estimation values and type estimation values of the scattering centers in turn; a scattering coefficient estimation value of each scattering center is determined based on the number, position estimation value and type estimation value of the scattering centers, and the scattering center features are extracted using a two-dimensional scattering center extraction technique.

[0061] Specifically, after SAR data has been range-corrected and phase-compensated, target resolution can be reduced to the problem of estimating the parameters of two-dimensional scattering centers. The complete process of the two-dimensional multiple signal classification (MUSIC) algorithm based on the geometric diffraction (GTD) model is as follows:

[0062] Step 1: Get the target "angle-frequency" two-dimensional original echo and perform Stolt interpolation on the echo;

[0063] Step 2 estimates the covariance matrix of the observed data;

[0064] Step 3: perform eigendecomposition on the covariance matrix and estimate the number of scattering centers using the Gaussian circle method;

[0065] Step 4 searches the defined MUSIC spectrum to obtain an estimate of the scattering center position;

[0066] Step 5: Perform a discrete search on the defined MUSIC spectrum to obtain an estimate of the scattering center type;

[0067] Step 6 obtains an estimate of the scattering coefficient of each scattering center.

[0068] Using two-dimensional scattering center extraction technology, simplified regularized SAR image reconstruction is adopted, and scattering center feature extraction processing is realized through the scattering center extraction method of watershed. Figure 6 shown.

[0069] Then, for step 102, a comparative analysis is performed on the latent layer feature information and the traditional feature information to determine a comparison relationship between the latent layer feature information and the traditional feature information.

[0070] In an embodiment of the present invention, the comparison between hidden layer features and traditional features includes visualizing the hidden layer feature information layer by layer according to the depth of the hierarchy of the feature extraction network to obtain a visualized image of the hidden layer feature information; and determining the traditional feature information corresponding to each hierarchy based on the visualized image.

[0071] Specifically, by visualizing the hidden layers of the network layer by layer, from the shallowest layers to the deepest layers, we can find feature correspondences that break the correlation. This mapping analysis requires the network to have good convergence and recognition accuracy. Therefore, forward analysis of the network of simulated civilian ship data is more suitable for effective feature comparison analysis.

[0072] First, we analyze the correlation between the shallow layer of the network and traditional features:

[0073] like Figure 7 The following figure shows the feature visualization results of the shallow layers 1 and 2 of the network, as well as the geometric characteristics of the target after geometric contour detection. Both layers exhibit edge and geometric shape characteristics, indicating a correlation between the shallow layers of the feature extraction network and the geometric features. Given that the hidden layer values of deep networks lack a normalized pattern, quantitative analysis using metrics such as similarity is currently unavailable. Addressing this issue requires more sophisticated hidden layer feature processing techniques.

[0074] Then analyze the correlation between the network middle layer and traditional features:

[0075] Texture information has macroscopic distribution characteristics and also characterizes the frequency of occurrence of local pixel pairs. It is highly correlated with energy, uniformity, etc. Figure 8 and Figure 9 The middle and shallow layers of the deep network shown represent macroscopic distribution characteristics at different energy (grayscale) levels. As the layers deepen, the feature representation tends to focus on local characteristics such as texture scalability. This indicates that the middle layers of the feature extraction network are correlated with texture features. Furthermore, these two hidden layer features alternate within the network: one layer of grayscale representation is often followed by two or three layers of local texture scalability. This cycle gradually transitions from macroscopic geometric structure to microscopic texture and scattering point scalability features as the network structure deepens.

[0076] Finally, we analyze the correlation between the underlying network and traditional features:

[0077] Depend on Figure 10 The visualization of the hidden layer features at the bottom layer of the network shows that the deeper the network goes, the more prominent the detailed features become. These prominent details have certain similarities with features such as GLCM texture and scattering centers. This indicates that the depth of the feature extraction network is correlated with texture and scattering center features. This similarity cannot be quantitatively analyzed with current research technology, but further research is needed.

[0078] With respect to step 104 , the key hidden layer features of the hidden layer feature information in the analysis result are determined according to preset requirements, and inverse activation mapping is performed on the key hidden layer features to obtain intelligent recognition distribution characteristics of the SAR image data.

[0079] After determining the correlation through the above steps, key hidden features can be selected from all hidden features according to actual application requirements to complete subsequent processing.

[0080] In the embodiment of the present invention, the reverse activation mapping is a method well known to those skilled in the art, such as Figure 11 As shown in the figure, through this mapping analysis method, it can be seen that the hidden layer features of the deep network have a certain correlation with the traditional recognition features. As the number of network layers deepens, the recognition features gradually transition from macroscopic (such as geometric contours, grayscale) to local details (such as texture, scattering center, etc.), thereby obtaining the intelligent recognition distribution characteristics of SAR image data.

[0081] It is worth noting that some features of the network's middle layer experience certain macro-micro fluctuations as the number of layers changes, and there are still some unknown hidden layer features that cannot be explained by traditional recognition features. The significance of these features cannot be effectively explained at the current scientific level.

[0082] Please refer to Figure 12 The embodiment of the present invention provides a SAR feature analysis device for intelligent network hidden layer mapping, the device comprising:

[0083] Extraction module S1, used to input the acquired SAR image data into a preset feature extraction network, and output hidden layer feature information and traditional feature information of the SAR image data;

[0084] An analysis module S2 is configured to compare and analyze the latent layer feature information and the traditional feature information to determine a correspondence relationship between the latent layer feature information and the traditional feature information;

[0085] The mapping module S3 is used to determine the key hidden layer features of the hidden layer feature information in the analysis result according to preset requirements, and perform inverse activation mapping on the key hidden layer features to obtain the intelligent recognition distribution characteristics of the SAR image data.

[0086] In the embodiment of the present invention, the traditional feature information includes geometric features, texture features and scattering center features.

[0087] In an embodiment of the present invention, the geometric features are extracted by: performing target coarse segmentation processing on the filtered original image according to constant false alarm rate; performing morphological filtering on the segmentation result in terms of area and distance to remove areas that do not meet preset requirements; and performing pixel multiplication calculation on the result image obtained by filtering and the original image to obtain the geometric features of the SAR image.

[0088] In an embodiment of the present invention, the scattering center features are extracted in the following manner: Stolt interpolation processing is performed on the angle-frequency two-dimensional original echo of the acquired SAR image data, and estimation calculation is performed on the processed data to obtain a covariance matrix; eigendecomposition is performed on the covariance matrix, and the decomposition result is estimated using the Gauss circle method to obtain the number of scattering centers; a preset two-dimensional multiple signal classification spectrum is searched and discretely searched respectively to obtain position estimation values and type estimation values of the scattering centers in turn; a scattering coefficient estimation value of each scattering center is determined based on the number, position estimation value, and type estimation value of the scattering centers, and the scattering center features are extracted using a two-dimensional scattering center extraction technology.

[0089] In an embodiment of the present invention, the comparative analysis of the hidden layer feature information and the traditional feature information to determine the corresponding relationship between the hidden layer feature information and the traditional feature information includes: visualizing the hidden layer feature information layer by layer according to the depth of the hierarchy of the feature extraction network to obtain a visualized image of the hidden layer feature information; and determining the traditional feature information corresponding to each hierarchy based on the visualized image.

[0090] In an embodiment of the present invention, determining the traditional feature information corresponding to each level based on the visual image includes: determining that the shallow layer of the feature extraction network is associated with the geometric features based on the edge features and geometric shape features of the visual image; determining that the middle layer of the feature extraction network is associated with the texture features based on the macroscopic distribution characteristics and scalability of the visual image; and determining that the deep layer of the feature extraction network is associated with the texture features and scattering center features based on the detail features of the visual image.

[0091] It should be noted that the SAR signature analysis device for intelligent network hidden layer mapping provided in the above embodiment is merely illustrated by the division of the aforementioned functional modules. In actual applications, the aforementioned functions can be assigned to different functional modules as needed, i.e., the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the SAR signature analysis device for intelligent network hidden layer mapping provided in the above embodiment and the SAR signature analysis method for intelligent network hidden layer mapping are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0092] The embodiment of the present application also provides a computer device, please refer to Figure 13The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the SAR feature analysis method for intelligent network hidden layer mapping provided by the above-mentioned method embodiments.

[0093] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set, or instruction set, which is loaded and executed by a processor to implement the SAR feature analysis method for intelligent network hidden layer mapping provided by the above-mentioned method embodiments.

[0094] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the SAR feature analysis method for intelligent network hidden layer mapping described in any of the above embodiments.

[0095] For the convenience of description, the above systems or devices are described as being divided into various modules or units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0096] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0097] Finally, it should be noted that, in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0098] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A SAR feature analysis method for intelligent network hidden layer mapping, characterized in that: The method comprises: Inputting the acquired SAR image data into a preset feature extraction network, and outputting hidden layer feature information and traditional feature information of the SAR image data; performing comparative analysis on the latent layer feature information and the traditional feature information to determine a comparison relationship between the latent layer feature information and the traditional feature information; The key hidden layer features of the hidden layer feature information in the analysis result are determined according to preset requirements, and the key hidden layer features are reversely activated mapped to obtain the intelligent recognition distribution characteristics of the SAR image data.

2. The method according to claim 1, wherein The traditional feature information includes geometric features, texture features and scattering center features.

3. The method according to claim 2, wherein The geometric features are extracted as follows: Perform target coarse segmentation processing on the filtered original image according to constant false alarm rate; Perform morphological filtering on the segmentation results based on area and distance to remove areas that do not meet the preset requirements; Perform pixel multiplication calculation on the result image obtained by filtering and the original image to obtain the geometric features of the SAR image.

4. The method according to claim 2, wherein The scattering center features are extracted in the following way: Performing Stolt interpolation processing on the angle-frequency two-dimensional original echo of the acquired SAR image data, and performing estimation calculation on the processed data to obtain a covariance matrix; Performing eigendecomposition on the covariance matrix and estimating the decomposition result using the Gaussian circle method to obtain the number of scattering centers; The preset two-dimensional multiple signal classification spectrum is searched and discretely searched respectively to obtain the position estimation value and type estimation value of the scattering center in turn; The scattering coefficient estimation value of each scattering center is determined according to the number, position estimation value and type estimation value of the scattering centers, and the scattering center features are extracted using a two-dimensional scattering center extraction technology.

5. The method according to claim 1, wherein The comparative analysis of the latent feature information and the traditional feature information to determine the correspondence relationship between the latent feature information and the traditional feature information includes: Performing layer-by-layer visualization processing on the hidden layer feature information according to the hierarchical depth of the feature extraction network to obtain a visualized image of the hidden layer feature information; Traditional feature information corresponding to each level is determined according to the visual image.

6. The method according to claim 5, wherein The determining, based on the visual image, the traditional feature information corresponding to each level includes: Determining, based on edge features and geometric shape features of the visualized image, that a shallow layer of the feature extraction network is associated with the geometric features; Determining, based on the macroscopic distribution characteristics and scalability of the visualized image, that a middle layer of the feature extraction network is associated with texture features; According to the detail features of the visual image, it is determined that the deep layer of the feature extraction network has correlation with the texture features and the scattering center features.

7. A SAR feature analysis device for intelligent network hidden layer mapping, characterized in that: The device comprises: An extraction module is used to input the acquired SAR image data into a preset feature extraction network, and output hidden layer feature information and traditional feature information of the SAR image data; An analysis module, configured to compare and analyze the latent layer feature information and the traditional feature information to determine a correspondence relationship between the latent layer feature information and the traditional feature information; The mapping module is used to determine the key hidden layer features of the hidden layer feature information in the analysis result according to preset requirements, and perform inverse activation mapping on the key hidden layer features to obtain the intelligent recognition distribution characteristics of the SAR image data.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images