SAR image adaptive processing method and system based on deep learning

Abstract

The embodiment of the invention discloses an SAR image adaptive processing method and system based on deep learning, and relates to the technical field of image processing. The method comprises the following steps: acquiring a to-be-processed SAR (Synthetic Aperture Radar) image; multi-scale features are extracted from the SAR image, the multi-scale features comprise low-frequency features and high-frequency features, the low-frequency features comprise large-scale structure information of the SAR image, and the high-frequency features comprise small-scale detail information of the SAR image; inputting the multi-scale features into a trained target deep neural network to obtain adaptive processing parameters corresponding to the SAR image; and processing the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a SAR image adaptive processing method and system based on deep learning. Background Art

[0002] Synthetic Aperture Radar (SAR) is a technology that uses the Doppler effect of radar waves to achieve high-resolution imaging of ground targets by processing continuous radar echo signals on a moving radar platform. SAR images have all-weather, all-day, and all-sky observation capabilities and are widely used in geological exploration, military reconnaissance, environmental monitoring and other fields.

[0003] However, SAR images also have some inherent defects, such as noise, scattering points, interference fringes, etc., which affect the quality and visibility of SAR images. Therefore, denoising and enhancing SAR images to improve the contrast and clarity of SAR images is one of the important tasks in SAR image processing.

[0004] Among the existing SAR image processing methods, most are based on traditional image processing techniques, such as filtering, enhancement, etc., which make it difficult to achieve effective processing of SAR images. At the same time, due to the complexity and diversity of SAR images, different images may require different processing methods and parameters. Therefore, how to adaptively adjust the processing methods and parameters according to different images is also a difficult problem in SAR image processing.

[0005] Based on this, it is necessary to study a method and system that can perform adaptive processing according to the actual situation of SAR images to improve the processing efficiency and accuracy of SAR images. Summary of the invention

[0006] This specification provides a SAR image adaptive processing method and system based on deep learning, which can automatically determine appropriate processing parameters according to the characteristics and content of the SAR image, realize adaptive denoising and enhancement of the SAR image, and thus improve the processing efficiency and accuracy of the SAR image. To achieve the above purpose, the technical solutions adopted in this specification are as follows: In one aspect, the present specification provides a SAR image adaptive processing method based on deep learning, the method comprising: Step 102, obtaining a SAR image to be processed; Step 104, extracting multi-scale features from the SAR image to be processed; the multi-scale features include low-frequency features and high-frequency features; the low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; Step 106, inputting the multi-scale features into a trained target deep neural network to obtain adaptive processing parameters corresponding to the SAR image; the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; Step 108: Process the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

[0007] On the other hand, the present specification provides a SAR image adaptive processing system based on deep learning, the system comprising: An acquisition module, used for acquiring the SAR image to be processed; A multi-scale feature extraction module, used to extract multi-scale features from the SAR image, wherein the multi-scale features include low-frequency features and high-frequency features, wherein the low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; The adaptive processing parameter determination module is used to input the multi-scale features into the trained target deep neural network to obtain the adaptive processing parameters corresponding to the SAR image, wherein the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; The image processing module is used to process the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

[0008] Based on the above technical solution, this specification can achieve the following technical effects: (1) Through the deep learning-based adaptive SAR image processing method and system provided in this specification, appropriate processing parameters can be automatically determined according to the characteristics and content of the SAR image, so as to achieve classification, denoising and enhancement of the SAR image, and improve the efficiency and accuracy of SAR image processing.

[0009] (2) In the SAR image adaptive processing method and system based on deep learning provided in this specification, by using feature descriptors to extract features of low-frequency components and high-frequency components in SAR images respectively, feature descriptions can be performed on low-frequency information and high-frequency information in SAR images respectively, so as to obtain large-scale structural information such as edges, textures, and shapes, and small-scale detail information such as scattering points, noise, and interference fringes, thereby making the feature extraction of SAR images richer and more complete, and having a certain degree of pertinence, thereby improving the recognizability and distinguishability of features in SAR images.

[0010] (3) In the SAR image adaptive processing method and system based on deep learning provided in this specification, by using the classification parameter as a constraint parameter in the process of determining the adaptive processing parameters corresponding to the SAR image to be processed, the corresponding denoising parameters and enhancement parameters can be determined according to the category of each area in the SAR image using the corresponding method, thereby making the determination process of the denoising parameters and enhancement parameters corresponding to each area more targeted.

[0011] (4) In the SAR image adaptive processing method and system based on deep learning provided in this specification, by querying the reference denoising parameters and reference enhancement parameters corresponding to the categories to which each area in the SAR image belongs from a pre-established processing parameter database according to the classification parameters, and then correcting the first denoising parameters and first enhancement parameters output by the target deep neural network according to the reference denoising parameters and reference enhancement parameters to obtain the second denoising parameters and second enhancement parameters, the accuracy and reliability of the denoising parameters and enhancement parameters can be improved, thereby improving the denoising effect and enhancement effect of the SAR image. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is an exemplary flow chart of a SAR image adaptive processing method based on deep learning shown in an embodiment of this specification.

[0013] Figure 2 It is a flowchart of the specific steps of the SAR image adaptive processing method based on deep learning shown in an embodiment of this specification. Figure 3 It is a structural schematic diagram of a SAR image adaptive processing system based on deep learning shown in an embodiment of this specification. Figure 4This is a schematic diagram of an exemplary application scenario of a deep learning-based SAR image adaptive processing system shown in an embodiment of this specification. DETAILED DESCRIPTION

[0014] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0015] It should be understood that the "system", "device", "unit" and / or "module" used in this specification is a method for distinguishing different components, elements, parts, parts or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0016] As shown in this specification and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "comprise" and "include" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0017] Flowcharts are used in this specification to illustrate the operations performed by the system according to the embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed precisely in order. Instead, the steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or one or more operations may be removed from these processes. Example 1

[0018] Please refer to Figure 1 , Figure 1 The figure is a flow chart of a SAR image adaptive processing method based on deep learning provided by this embodiment. The method comprises: Step 102, obtaining a SAR image to be processed; It should be noted that in the embodiments of this specification, the SAR image to be processed may include various types and sources. For example, in some embodiments, the SAR image may be obtained by a SAR sensor mounted on an aerial platform, and exemplary aerial platforms include fixed-wing aircraft, helicopters, or drones. In some embodiments, the SAR image may be obtained by a SAR sensor mounted on a satellite, and the SAR image may be used to display an image of the earth's surface. In some embodiments, the SAR image may also be obtained through an open data set, for example, by accessing the data distribution platform of the ESA (European Space Agency).

[0019] Step 104, extracting multi-scale features from the SAR image to be processed; the multi-scale features include low-frequency features and high-frequency features; the low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; In this embodiment, one implementation of step 104 is: Step 202, performing low-pass filtering on the SAR image based on a Gaussian filter to obtain a low-frequency component in the SAR image; Specifically, the SAR image to be processed can be low-pass filtered by a Gaussian filter to obtain the low-frequency component in the SAR image. The low-frequency component can contain large-scale structural information of the SAR image to be processed, such as edges, textures, shapes, etc. By screening out the low-frequency component of the SAR image, the high-frequency noise and scattering points in the SAR image can be removed, and only the low-frequency information in the SAR image is retained, thereby providing a data basis for subsequent low-frequency feature extraction.

[0020] Step 204, subtracting the low-frequency component from the SAR image to obtain a high-frequency component of the SAR image; Specifically, the aforementioned low-frequency component can be subtracted from the SAR image to obtain the high-frequency component of the SAR image to be processed. The high-frequency component may contain small-scale detail information of the SAR image to be processed, such as scattering points, noise, interference fringes, etc. By screening out the high-frequency component of the SAR image, the high-frequency information in the SAR image can be highlighted, providing a data basis for subsequent high-frequency feature extraction.

[0021] It should be noted that the aforementioned low-frequency components and high-frequency components refer to information in different frequency ranges in SAR images. Generally speaking, low-frequency components contain large-scale structural information of SAR images, such as edges, textures, shapes, etc., while high-frequency components contain small-scale detail information of SAR images, such as scattering points, noise, interference fringes, etc. Among them, scattering points refer to points in SAR images that have strong reflection signals due to the unevenness of the target surface or volume. They usually appear as bright or dark spots in SAR images, which can reflect the physical characteristics of the target, such as shape, material, direction, etc., but will also affect the quality and visibility of SAR images, so denoising is required. Interference fringes refer to periodic bright and dark fringes in SAR images due to coherent interference. They usually appear as colored fringes in SAR images, which can reflect the elevation changes of the target, such as terrain, landforms, earthquakes, etc., but will also affect the contrast and clarity of SAR images, so enhancement processing is required.

[0022] Step 206, using a feature descriptor to extract features of the low-frequency component and the high-frequency component respectively, to obtain low-frequency features and high-frequency features in the SAR image; the feature descriptor includes any one or any combination of SIFT, SURF, and HOG; Specifically, after obtaining the low-frequency component and the high-frequency component of the SAR image to be processed through the aforementioned steps, the feature descriptor can be used to extract the features of the aforementioned low-frequency component to obtain the low-frequency features in the SAR image to be processed. Similarly, the feature descriptor can be used to extract the features of the aforementioned high-frequency component to obtain the high-frequency features in the SAR image to be processed. In the embodiments of this specification, the aforementioned low-frequency features may refer to the large-scale structural information of the SAR image, such as edges, textures, shapes, etc.; the high-frequency features may refer to the small-scale detail information of the SAR image, such as scattering points, noise, interference fringes, etc.

[0023] In this embodiment, the aforementioned feature descriptors may include any one of SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), HOG (Histogram of Oriented Gradients), or a combination thereof, which may be used to extract key points, directions, gradients, and other information in SAR images.

[0024] For example, the SIFT detector can be used for the aforementioned low-frequency component and high-frequency component respectively. The SIFT detector can detect stable key points in the SAR image at different scales and determine the direction of each key point. For each key point, the image area around it can be extracted according to the direction and scale detected. Then, for the area, the corresponding gradient histogram can be calculated. The histogram can represent the gradient distribution in different directions in the area. Through the above operations, a multi-dimensional SIFT feature vector of each key point can be obtained, wherein each dimension corresponds to a bin in the gradient histogram. In the embodiment of this specification, the multi-dimensional SIFT feature vector obtained above can be used to represent the low-frequency features and high-frequency features of the SAR image. Finally, by integrating or splicing the corresponding SIFT feature vectors of the low-frequency component and the high-frequency component, a multi-scale feature representation of the SAR image can be obtained. Through the multi-scale feature, not only can the structure, texture, edge, scattering point, noise, interference fringes and other information in the SAR image be reflected at the same time, but also it is invariant to scale and rotation changes, so as to better describe and analyze the target features in the SAR image to be processed.

[0025] For example, the Hessian matrix can also be used for the aforementioned low-frequency components and high-frequency components to obtain key points and directions, wherein the Hessian matrix is ​​a mathematical tool for image processing, which can find local extreme points in the image at different scales and determine the direction of each key point. By calculating the Hessian matrix for the aforementioned low-frequency components and high-frequency components, the position and direction information of the key points in the SAR image can be obtained. Furthermore, for each key point, the image area around it can be extracted according to its direction and scale, and the Haar wavelet transform of the area can be calculated to obtain the characteristic response of the area to form a multi-dimensional SURF feature vector. It can be understood that by extracting the Haar wavelet transform of the surrounding image area for each key point, a multi-dimensional SURF feature vector can be obtained, wherein each dimension corresponds to the calculated characteristic response value. Finally, based on the multi-dimensional SURF feature vector, the low-frequency features and high-frequency features of the SAR image can be obtained. It should be pointed out that by using the Hessian matrix to detect key points and directions and combining it with the Haar wavelet transform to extract feature vectors, image features can be detected and described at different scales, and a multi-scale feature representation of the SAR image can be obtained, thereby more accurately describing and analyzing the target features in the image.

[0026] For example, the amplitude and direction of the gradient can be calculated for the aforementioned low-frequency component and high-frequency component, respectively, and the aforementioned low-frequency component and high-frequency component can be divided into several units, respectively, and then the direction histogram of the gradient can be calculated in each unit. Specifically, the low-frequency component and the high-frequency component can be divided into multiple units (such as pixel blocks or grid units), respectively, and the histogram of the gradient direction can be calculated in each unit. The histogram can count the frequency of occurrence of different gradient directions in each unit, thereby reflecting the structure, edge, texture, shape and other features in the low-frequency component, as well as the scattering points, noise, interference fringes and other features in the high-frequency component. In some embodiments, n adjacent (for example, n=4, n=9) units can be grouped into a block, and the histogram in each block can be normalized (also called color space standardization) to form a HOG feature vector. It should be pointed out that through normalization, the feature vector can be made robust to image changes such as illumination and contrast. Further, based on the aforementioned HOG feature vector, the low-frequency features and high-frequency features of the SAR image can be obtained.

[0027] It should also be pointed out that in some embodiments of the present specification, appropriate feature descriptors such as SIFT, SURF, HOG, etc. can be selected according to different application scenarios and requirements to adapt to different feature extraction targets such as key points, directions, gradients, etc., thereby improving the flexibility and efficiency of feature extraction of SAR images.

[0028] In the embodiments of the present specification, by using feature descriptors to extract features of the aforementioned low-frequency components and high-frequency components respectively, feature descriptions can be performed on the low-frequency information and high-frequency information in the SAR image respectively, and large-scale structural information such as edges, textures, and shapes and small-scale detail information such as scattering points, noise, and interference fringes can be obtained, thereby making the feature extraction of the SAR image richer and more complete, and having a certain degree of pertinence, thereby improving the recognizability and distinguishability of the features in the SAR image.

[0029] Step 208: splice the low-frequency features and the high-frequency features to obtain multi-scale features of the SAR image.

[0030] In this embodiment, step 104 may also use a convolutional neural network (CNN) or a residual network (ResNet) to extract features from the aforementioned SAR image to obtain multi-scale features including the aforementioned low-frequency features and high-frequency features. Exemplarily, in some embodiments, the SAR image may be used as input, and processed by multiple convolutional layers and pooling layers to obtain feature maps of different scales, and then the feature map may be average pooled or processed by a fully connected layer to obtain a feature vector as a multi-scale feature of the SAR image. This method can automatically learn the features of the SAR image without artificially designing feature descriptors, but requires a large amount of annotated data and computing resources for training. For example, multiple sample images annotated with multi-scale features may be used as training data to train the initial convolutional neural network (CNN) or residual network (ResNet) so that it acquires the ability to extract multi-scale features from SAR images.

[0031] Step 106, inputting the multi-scale features into a trained target deep neural network to obtain adaptive processing parameters corresponding to the SAR image; the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; Specifically, the target deep neural network can be a multilayer perceptron (MLP), which includes an input layer, a hidden layer and an output layer.

[0032] The input layer can be used to receive the multi-scale features of the SAR image extracted in the above steps as the input of the network. As can be seen from the above, the multi-scale features can include low-frequency features and high-frequency features, which can respectively reflect the large-scale structural information and small-scale detail information of the SAR image.

[0033] The hidden layer may include several fully connected layers, each of which is composed of several neurons. Each neuron can perform a weighted summation on the input features and output an activation value through a nonlinear activation function (such as ReLU function, Sigmoid function or Tanh function) as the input of the next layer.

[0034] The output layer can include three fully connected layers, corresponding to classification parameters, denoising parameters and enhancement parameters respectively. Each fully connected layer can be composed of several neurons. Each neuron can perform weighted summation on the input features and output a prediction value as the output of the network through a suitable activation function (such as Softmax function, Sigmoid function or linear function).

[0035] In this embodiment, the target deep neural network can realize prediction and learning through two processes of forward propagation and back propagation. Specifically: forward propagation refers to transferring the multi-scale features of the SAR image from the input layer to the output layer layer by layer to obtain the predicted value of the adaptive processing parameter corresponding to the SAR image. This process is determined by the network parameters (such as weights and biases) of the network, and the network parameters can be used to control the output of the network; back propagation refers to transferring the distance (or loss) between the predicted value of the output layer and the sample adaptive processing parameter corresponding to the training data from the output layer to the input layer layer by layer to obtain the gradient of the network parameters. This process is determined by the activation function and loss function of the network, which are the hyperparameters of the network and are used to control the learning of the network.

[0036] In this embodiment, the target deep neural network can be trained based on the following method: First, initialize the network parameters, such as weights and biases, using random numbers or other methods.

[0037] Then, multiple groups of sample SAR images are used as training data to input into the network, wherein each sample SAR image is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample multi-scale features can reflect the features of the sample SAR image, and the sample adaptive processing parameters can reflect the processing method and degree of the sample SAR image. In the embodiment of this specification, the sample multi-scale features can include low-frequency features for reflecting large-scale structural information and high-frequency features for reflecting small-scale detail information, and the sample adaptive processing parameters can include classification parameters, denoising parameters and enhancement parameters. It should be noted that in the embodiment of this specification, the aforementioned classification parameters refer to parameters for characterizing the categories of each region in the SAR image. For example, the SAR image can be divided into several regions according to the pixel value, texture feature, type of ground object and other information in the SAR image, each region corresponds to a category (such as water area, vegetation, building, etc.), and each category can correspond to a parameter. Denoising parameters refer to parameters for controlling the denoising method, degree and effect, such as the type, size, threshold and the like of the filter. In the embodiment of this specification, the denoising parameters can also correspond to various processing parameters in the wavelet threshold denoising method used in the subsequent process, such as the target threshold function involved in the subsequent process. Enhancement parameters refer to parameters used to control the enhancement method, degree and effect, such as equalization parameters. Specifically, the equalization parameters may include contrast adjustment parameters, clarity adjustment parameters, brightness adjustment parameters, etc. In some embodiments, the enhancement parameters may correspond to the target equalization method in the subsequent process.

[0038] Then, for each training data, forward propagation and back propagation can be performed to obtain the network output and gradient, and the gradient descent method or other optimization methods can be used to update the network parameters according to the gradient, so that the network output is closer to the actual parameters of the training data (such as the aforementioned sample adaptive processing parameters).

[0039] Finally, the above steps are repeated until the distance between the output of the network and the sample adaptive processing parameter corresponding to the training data is less than or equal to the preset threshold, or the preset number of training rounds is reached, and the target deep neural network is obtained. The preset threshold or the preset number of training rounds can be set according to experience or actual needs, and therefore, it is not specifically limited in this specification.

[0040] After the target deep neural network is trained through the above steps, the multi-scale features obtained in step 104 are input into the trained target deep neural network, and the adaptive processing parameters corresponding to the aforementioned SAR image to be processed can be obtained. It can be understood that since the target neural network is trained based on the sample multi-scale features and sample adaptive processing parameters annotated on the sample SAR image, the adaptive processing parameters output by the target deep neural network can correspond to the multi-scale features of the SAR image to be processed. Specifically, the adaptive processing parameters include classification parameters, denoising parameters and enhancement parameters. Among them, the classification parameters refer to parameters used to characterize the categories of each region in the SAR image. For example, the SAR image can be divided into several regions according to the pixel value, texture feature, type of ground object and other information in the SAR image, each region corresponds to a category (such as water area, vegetation, building, farmland, road, etc.), and each category can correspond to a parameter. Denoising parameters refer to parameters used to control the denoising method, degree and effect, such as the type, size, threshold and the like of the filter. In the embodiment of this specification, the denoising parameters can also correspond to the various processing parameters in the wavelet threshold denoising method used in the subsequent process, such as the target threshold function involved in the subsequent process. Enhancement parameters refer to parameters used to control the enhancement method, degree and effect, such as equalization parameters. Specifically, the equalization parameters may include contrast adjustment parameters, clarity adjustment parameters, brightness adjustment parameters, etc. In some embodiments, the enhancement parameters may correspond to the target equalization method in the subsequent process.

[0041] In this embodiment, the SAR image to be processed may be divided into multiple (eg, two or more) regions according to the classification parameters, wherein each region may correspond to a set of adaptive processing parameters.

[0042] It should be noted that in the embodiments of this specification, the aforementioned classification parameter can be used as a constraint parameter. Specifically, when the target deep neural network determines the adaptive processing parameters corresponding to the SAR image to be processed, it can use corresponding methods to determine the corresponding denoising parameters and enhancement parameters according to the categories of each area in the SAR image, so that the determination process of the denoising parameters and enhancement parameters corresponding to each area is more targeted.

[0043] Step 108: Process the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

[0044] In this embodiment, the adaptive processing parameters corresponding to the SAR image include a classification parameter, a first denoising parameter and a first enhancement parameter; In this embodiment, the first denoising parameter and the first enhancement parameter may be directly used to perform SAR image processing, and the steps include: In terms of denoising, the image processing module can perform wavelet transform on the SAR image to be processed to obtain the wavelet coefficients of the SAR image, and then perform threshold processing on the wavelet coefficients according to the first denoising parameter obtained by the target deep neural network. Further, by performing inverse wavelet transform on the wavelet coefficients after the threshold processing, the denoised SAR image can be obtained.

[0045] In this embodiment, the process of performing threshold processing on the wavelet coefficient may include: S101, selecting a target threshold function according to a first denoising parameter, wherein the target threshold function may include a hard threshold function, a soft threshold function or an improved threshold function.

[0046] S102, calculating the absolute value of each component of the wavelet coefficients, wherein each component of the wavelet coefficients refers to a set of values ​​obtained after wavelet transformation, which represents the components of the signal at different frequencies or scales, and can reflect the local characteristics of the signal.

[0047] S103, comparing the absolute value with the target threshold function, and performing value assignment processing on the wavelet coefficient according to the comparison result and the target threshold function to obtain processed wavelet coefficients. The value assignment processing includes keeping unchanged, setting to zero, and subtracting a threshold.

[0048] In this embodiment, by performing wavelet transform on the SAR image and performing threshold processing on the wavelet coefficient according to the first denoising parameter obtained by the target deep neural network, the characteristics of the wavelet transform can be utilized to concentrate the energy of the signal in some large wavelet coefficients, while the energy of the noise is dispersed in the entire wavelet domain. Therefore, by setting a threshold, the wavelet coefficients smaller than the threshold are regarded as noise and removed, thereby achieving the purpose of noise reduction.

[0049] The target threshold function refers to a function used to process wavelet coefficients, which defines how the wavelet coefficients should be assigned when the absolute value of the wavelet coefficients is greater than or less than the threshold value, so as to retain the main information of the signal while removing the interference of noise. In an embodiment of the present specification, there may be a mapping relationship between the aforementioned first denoising parameter and the target threshold function. After the first denoising parameter is obtained through the aforementioned steps, the target threshold function can be determined through a pre-configured mapping relationship. In some embodiments, the same target threshold function may correspond to a denoising parameter range. Exemplarily, in an embodiment of the present specification, the target threshold function may include a hard threshold function, a soft threshold function, or an improved threshold function. Wherein: The rule corresponding to the hard threshold function is: when the absolute value of the wavelet coefficient is less than the threshold, it is set to 0; when the absolute value of the wavelet coefficient is greater than or equal to the threshold, its original value remains unchanged. For example, if the threshold is 2, then when the wavelet coefficient is 1.5, it becomes 0 after being processed by the hard threshold function; when the wavelet coefficient is 2.5, it remains 2.5 after being processed by the hard threshold function.

[0050] The rule corresponding to the soft threshold function is that when the absolute value of the wavelet coefficient is less than the threshold, it is set to 0. When the absolute value of the wavelet coefficient is greater than the threshold, it is subtracted from the threshold and its original sign is maintained. For example, if the threshold is 2, then when the wavelet coefficient is 1.5, it becomes 0 after being processed by the soft threshold function; when the wavelet coefficient is 2.5, it becomes 0 after being processed by the soft threshold function, and the sign remains unchanged.

[0051] The improved threshold function refers to making some modifications or optimizations on the basis of the hard threshold function and the soft threshold function to improve the denoising effect and reduce the generation of distortion or artifacts. As an example, in an improved threshold function, the rule is: when the absolute value of the wavelet coefficient is less than the threshold, it is set to 0; when the absolute value of the wavelet coefficient is greater than the threshold, it is multiplied by a coefficient and its original sign is maintained. This method can prevent large wavelet coefficients from being over-compressed, thereby retaining more signal details.

[0052] Furthermore, with regard to enhancement processing, the image processing module can perform histogram statistics on the SAR image to be processed to obtain the grayscale distribution of the SAR image, and then equalize the grayscale distribution according to the first enhancement parameter output by the target deep neural network. Finally, by performing inverse mapping on the equalized grayscale distribution, an enhanced SAR image is obtained.

[0053] In this embodiment, the process of equalizing the grayscale distribution according to the first enhancement parameter may include: S111, selecting a target equalization method according to the first enhancement parameter, wherein the target equalization method includes global equalization, local equalization, and adaptive equalization.

[0054] S112: transform the grayscale distribution according to the target equalization method to obtain a equalized grayscale distribution.

[0055] In this embodiment, the mapping relationship between the enhancement parameter and the target equalization method can be preconfigured. In some embodiments, the same target equalization method can correspond to a range of enhancement parameters. After obtaining the first enhancement parameter through the above steps, the image processing module can determine the target equalization method based on the mapping relationship between the enhancement parameter and the target equalization method. Among them, the target equalization method can include global equalization, local equalization, and adaptive equalization.

[0056] After determining the target equalization method, the image processing module can transform the grayscale distribution of the SAR image to be processed according to the target equalization method to adjust image parameters such as contrast, brightness and clarity of the SAR image to achieve grayscale distribution equalization.

[0057] In this embodiment, the process of performing inverse mapping on the equalized grayscale distribution to obtain an enhanced SAR image may include: S121, according to the equalized grayscale distribution, calculate the cumulative distribution function corresponding to each grayscale level. The cumulative distribution function (CDF) refers to the proportion of each grayscale level and the grayscale levels below it. In some embodiments, the cumulative distribution function can be expressed as follows:

[0058] Where C(k) is the value of the cumulative distribution function, p(i) is the probability of occurrence of a pixel with gray level i, n(i) is the number of pixels with gray level i, M and N are the number of rows and columns of the SAR image, k is the range of gray levels, usually 0 to L−1, and L is the total number of gray levels, usually 256.

[0059] S122, according to the cumulative distribution function, each gray level is mapped to a new gray level, so that the difference between the new gray level and the original gray level is minimized, while the gray level distribution after equalization is kept unchanged. In step S122, by mapping each gray level to a new gray level according to the cumulative distribution function, so that the difference between the new gray level and the original gray level is minimized, while the gray level distribution after equalization is kept unchanged, the color distortion or change caused by histogram equalization can be avoided, and the naturalness and authenticity of the image can be improved. In some embodiments, the nearest neighbor interpolation method can be used in the aforementioned gray level mapping method, that is, finding the value closest to the original gray level in the cumulative distribution function, and then using the corresponding gray level as the new gray level. The mapping formula can be expressed as follows:

[0060] Among them, s is the new gray level, r is the original gray level, T is the mapping function, C is the cumulative distribution function, and k is the range of gray levels.

[0061] S123, according to the new grayscale, replacing each pixel value of the SAR image with a corresponding new grayscale value to obtain an enhanced SAR image.

[0062] After determining the new grayscale through the above steps, the image processing module may replace each pixel value of the SAR image with the corresponding new grayscale value, thereby obtaining an enhanced SAR image.

[0063] In this embodiment, reference Figure 2 , another implementation of step 108 is: Step 302, querying reference denoising parameters and reference enhancement parameters corresponding to the category to which the SAR image belongs from a pre-established processing parameter database according to the classification parameters; Step 304, performing comprehensive processing according to the reference denoising parameter and the first denoising parameter to obtain a second denoising parameter; Step 306, performing comprehensive processing according to the reference enhancement parameter and the first enhancement parameter to obtain a second enhancement parameter; Step 308: Perform denoising and enhancement processing on the SAR image based on the second denoising parameter and the second enhancement parameter to obtain a processed SAR image.

[0064] In this embodiment, step 308 includes: Step 402, performing wavelet transform on the SAR image to obtain wavelet coefficients of the SAR image; Step 404, performing threshold processing on the wavelet coefficients according to the second denoising parameter to obtain wavelet coefficients after threshold processing; In this embodiment, one implementation of step 404 is: Step 4041, selecting a target threshold function according to the second denoising parameter; the target threshold function includes a hard threshold function, a soft threshold function or an improved threshold function; Step 4042, calculating the absolute value of each component of the wavelet coefficient to obtain the component absolute value; Step 4043, comparing the component absolute value with the target threshold function to obtain a comparison result; Step 4044, assigning values ​​to the wavelet coefficients according to the comparison result and the target threshold function to obtain processed wavelet coefficients; the assignment process includes keeping unchanged, setting to zero, and subtracting a threshold.

[0065] Step 406: Perform inverse wavelet transform on the wavelet coefficients after the threshold processing to obtain a denoised SAR image.

[0066] In this embodiment, step 308 further includes: Step 502, performing histogram statistics on the SAR image to obtain the grayscale distribution of the SAR image; Step 504, equalizing the grayscale distribution according to the second enhancement parameter to obtain a equalized grayscale distribution; In this embodiment, one implementation of step 504 is: Step 5041: selecting a target equalization method according to the second enhancement parameter; the target equalization method includes global equalization, local equalization, and adaptive equalization; Step 5042: transform the grayscale distribution according to the target equalization method to obtain a equalized grayscale distribution.

[0067] Step 506: De-map the equalized grayscale distribution to obtain an enhanced SAR image.

[0068] In this embodiment, one implementation of step 506 is: Step 5061, calculating the cumulative distribution function corresponding to each gray level according to the equalized gray distribution; Step 5062, mapping each gray level to a new gray level according to the cumulative distribution function, so that the difference between the new gray level and the original gray level is minimized, while keeping the equalized gray level distribution unchanged; Step 5063: According to the new grayscale, each pixel value of the SAR image is replaced with a corresponding new grayscale value to obtain an enhanced SAR image.

[0069] Specifically, in order to ensure the accuracy and reliability of the first denoising parameter and the first enhancement parameter output by the target deep neural network and avoid adding the step of correcting the first denoising parameter and the first enhancement parameter, the following method is adopted.

[0070] In this embodiment, reference denoising parameters and reference enhancement parameters corresponding to the categories to which each region in the SAR image belongs can be queried from a pre-established processing parameter database according to the classification parameters obtained in the aforementioned steps. Then, the reference denoising parameters and the aforementioned first denoising parameters are comprehensively processed to obtain second denoising parameters, and the reference enhancement parameters and the aforementioned first enhancement parameters are comprehensively processed to obtain second enhancement parameters.

[0071] In this embodiment, the difference between the first denoising parameter and the reference denoising parameter can be calculated, wherein the first denoising parameter and the reference denoising parameter can be represented by a vector, and the difference between the first denoising parameter and the reference denoising parameter can point to the distance between quantities (for example, Euclidean distance or Manhattan distance). In some embodiments, when the distance between the first denoising parameter and the reference denoising parameter is less than or equal to a first threshold, the first denoising parameter can be directly used as the second denoising parameter; when the distance between the first denoising parameter and the reference denoising parameter is greater than the first threshold and less than the second threshold (the second threshold is greater than the first threshold), the mean of the first denoising parameter and the reference denoising parameter can be used as the second denoising parameter; when the distance between the first denoising parameter and the reference denoising parameter is greater than or equal to the second threshold, the reference denoising parameter can be used as the second denoising parameter.

[0072] Similarly, when the distance between the first enhancement parameter and the reference enhancement parameter is less than or equal to a first threshold, the first enhancement parameter can be directly used as the above-mentioned second enhancement parameter; when the distance between the first enhancement parameter and the reference enhancement parameter is greater than the first threshold and less than the second threshold (wherein the second threshold is greater than the first threshold), the average of the first enhancement parameter and the reference enhancement parameter can be used as the above-mentioned second enhancement parameter; when the distance between the first enhancement parameter and the reference enhancement parameter is greater than or equal to the second threshold, the reference enhancement parameter can be used as the above-mentioned second enhancement parameter.

[0073] It should be noted that in the embodiments of this specification, the first threshold and the second threshold can be set according to experience or actual needs, and are not specifically limited in this specification.

[0074] After the denoising parameters and the enhancement parameters are corrected through the above correction process, the image processing module can perform denoising and enhancement processing on the SAR image based on the second denoising parameters and the second enhancement parameters obtained after correction to obtain a processed SAR image.

[0075] Specifically, in terms of denoising, the image processing module can perform wavelet transform on the SAR image to be processed to obtain wavelet coefficients of the SAR image, and then perform threshold processing on the wavelet coefficients according to the aforementioned second denoising parameter. Further, by performing inverse wavelet transform on the wavelet coefficients after the threshold processing, the denoised SAR image can be obtained.

[0076] Furthermore, with regard to enhancement processing, the image processing module 240 may perform histogram statistics on the SAR image to be processed to obtain the grayscale distribution of the SAR image, and then equalize the grayscale distribution according to the aforementioned second enhancement parameter, and finally, obtain the enhanced SAR image by performing inverse mapping on the equalized grayscale distribution.

[0077] It can be understood that the second denoising parameter and the second enhancement parameter undergo a correction step in the determination process, and therefore, they have higher accuracy and reliability than the first denoising parameter and the first enhancement parameter. Accordingly, the SAR image obtained by denoising and enhancing by the second denoising parameter and the second enhancement parameter can have better denoising and enhancement effects.

[0078] In the embodiment of the present specification, more details of denoising and enhancing the SAR image based on the second denoising parameter and the second enhancement parameter can refer to the process of denoising and enhancing the SAR image based on the first denoising parameter and the first enhancement parameter described above, and therefore, they will not be described in detail here.

[0079] Based on this, the beneficial effects that may be brought about by this embodiment include but are not limited to: (1) This method can automatically determine appropriate processing parameters according to the characteristics and content of SAR images, realize the classification, denoising and enhancement of SAR images, and improve the efficiency and accuracy of SAR image processing.

[0080] (2) This method uses feature descriptors to extract features from low-frequency components and high-frequency components in SAR images respectively. It can describe the features of low-frequency information and high-frequency information in SAR images respectively, and obtain large-scale structural information such as edges, textures, and shapes, and small-scale detail information such as scattering points, noise, and interference fringes. This makes the feature extraction of SAR images richer and more complete, and has a certain degree of specificity, thereby improving the recognizability and distinguishability of features in SAR images.

[0081] (3) This method uses the classification parameter as a constraint parameter in the process of determining the adaptive processing parameters corresponding to the SAR image to be processed. The corresponding method can be used to determine the corresponding denoising parameters and enhancement parameters of each area in the SAR image according to the category of each area, thereby making the determination process of the denoising parameters and enhancement parameters corresponding to each area more targeted.

[0082] (4) This method queries the reference denoising parameters and reference enhancement parameters corresponding to the categories of each area in the SAR image from a pre-established processing parameter database according to the classification parameters, and then corrects the first denoising parameters and first enhancement parameters output by the target deep neural network according to the reference denoising parameters and reference enhancement parameters to obtain the second denoising parameters and second enhancement parameters, which can improve the accuracy and reliability of the denoising parameters and enhancement parameters, thereby improving the denoising effect and enhancement effect of the SAR image. Example 2

[0083] refer to Figure 3 , Figure 3 The figure shows a schematic diagram of the structure of a SAR image adaptive processing system based on deep learning provided by this embodiment. The system includes: An acquisition module 210 is used to acquire a SAR image to be processed; A multi-scale feature extraction module 220 is used to extract multi-scale features from the SAR image, wherein the multi-scale features include low-frequency features and high-frequency features, wherein the low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; The adaptive processing parameter determination module 230 is used to input the multi-scale features into the trained target deep neural network to obtain the adaptive processing parameters corresponding to the SAR image, wherein the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; The image processing module 240 is used to process the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

[0084] Optionally, the multi-scale feature extraction module includes: A low-frequency component extraction submodule, used for performing low-pass filtering on the SAR image based on a Gaussian filter to obtain a low-frequency component in the SAR image; A high frequency component extraction submodule, used for subtracting the low frequency component from the SAR image to obtain the high frequency component of the SAR image; A feature extraction submodule, used to use a feature descriptor to perform feature extraction on the low-frequency component and the high-frequency component respectively, so as to obtain low-frequency features and high-frequency features in the SAR image; the feature descriptor includes any one or any combination of SIFT, SURF, and HOG; The feature splicing submodule is used to splice the low-frequency features and the high-frequency features to obtain multi-scale features of the SAR image.

[0085] Optionally, the large-scale structural information includes edges, textures, and shapes, and the small-scale detail information includes scattering points, noise, and interference fringes.

[0086] Optionally, the adaptive processing parameters corresponding to the SAR image include a classification parameter, a first denoising parameter and a first enhancement parameter; Optionally, the image processing module includes: A reference parameter determination submodule, used to query reference denoising parameters and reference enhancement parameters corresponding to the category to which the SAR image belongs from a pre-established processing parameter database according to the classification parameters; A second denoising parameter determination submodule, configured to perform comprehensive processing based on the reference denoising parameter and the first denoising parameter to obtain a second denoising parameter; A second enhancement parameter determination submodule, configured to perform comprehensive processing according to the reference enhancement parameter and the first enhancement parameter to obtain a second enhancement parameter; The image processing submodule is used to perform denoising and enhancement processing on the SAR image based on the second denoising parameter and the second enhancement parameter to obtain a processed SAR image.

[0087] Optionally, the image processing submodule includes: A wavelet coefficient determination unit, used for performing wavelet transform on the SAR image to obtain the wavelet coefficient of the SAR image; A threshold processing unit, configured to perform threshold processing on the wavelet coefficients according to the second denoising parameter to obtain the wavelet coefficients after the threshold processing; The denoising processing unit is used to perform inverse wavelet transform on the wavelet coefficients after the threshold processing to obtain a denoised SAR image.

[0088] Optionally, the threshold processing unit includes: A target threshold function determination subunit, configured to select a target threshold function according to the second denoising parameter; the target threshold function includes a hard threshold function, a soft threshold function or an improved threshold function; An absolute value calculation subunit, used for calculating the absolute value of each component of the wavelet coefficient to obtain the absolute value of the component; A numerical comparison subunit, used for comparing the absolute value of the component with the target threshold function to obtain a comparison result; The value assignment processing subunit is used to perform value assignment processing on the wavelet coefficients according to the comparison result and the target threshold function to obtain processed wavelet coefficients; the value assignment processing includes keeping unchanged, setting to zero, and subtracting a threshold.

[0089] Optionally, the image processing module further includes: A histogram statistics unit, used for performing histogram statistics on the SAR image to obtain the grayscale distribution of the SAR image; A grayscale equalization unit, configured to equalize the grayscale distribution according to the second enhancement parameter to obtain a equalized grayscale distribution; The enhancement processing unit is used to perform inverse mapping on the equalized grayscale distribution to obtain an enhanced SAR image.

[0090] Optionally, the grayscale equalization unit includes: A target equalization method determination subunit, configured to select a target equalization method according to the second enhancement parameter; the target equalization method includes global equalization, local equalization, and adaptive equalization; The grayscale transformation subunit is used to transform the grayscale distribution according to the target equalization method to obtain a balanced grayscale distribution.

[0091] Optionally, the enhanced processing unit includes: A cumulative distribution function determination subunit, used to calculate the cumulative distribution function corresponding to each gray level according to the equalized gray distribution; A new grayscale determination subunit, configured to map each grayscale to a new grayscale according to the cumulative distribution function, so that the difference between the new grayscale and the original grayscale is minimized, while keeping the equalized grayscale distribution unchanged; The grayscale replacement subunit is used to replace each pixel value of the SAR image with a corresponding new grayscale value according to the new grayscale level to obtain an enhanced SAR image.

[0092] It should be understood that Figure 3 The deep learning-based SAR image adaptive processing system 200 and its modules shown can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art can understand that the above-mentioned methods and systems can be implemented using computer executable instructions and / or included in processor control codes, such as such codes provided on carrier media such as disks, CDs or DVD-ROMs, programmable memories such as read-only memories (firmware), or data carriers such as optical or electronic signal carriers. The system and its modules of this specification can not only be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., but can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software (for example, firmware).

[0093] It should be noted that the above description of the SAR image adaptive processing system 200 based on deep learning is provided for illustrative purposes only and is not intended to limit the scope of this specification. It is understandable that for those skilled in the art, according to the description of this specification, without departing from this principle, the various modules can be arbitrarily combined, or a subsystem can be formed to connect with other modules. For example, Figure 2 The acquisition module 210, multi-scale feature extraction module 220, adaptive processing parameter determination module 230 and image processing module described in the above can be different modules in a system, or one module can realize the functions of two or more of the above modules. In some embodiments, the above modules can be part of the processing device 130 and / or the terminal device 140.

[0094] Based on this, the beneficial effects that may be brought about by this embodiment include but are not limited to: (1) This system can automatically determine appropriate processing parameters based on the characteristics and content of SAR images, realize classification, denoising and enhancement of SAR images, and improve the efficiency and accuracy of SAR image processing.

[0095] (2) This system uses feature descriptors to extract features from low-frequency components and high-frequency components in SAR images respectively. It can describe the features of low-frequency information and high-frequency information in SAR images respectively, and obtain large-scale structural information such as edges, textures, and shapes, and small-scale detail information such as scattering points, noise, and interference fringes. This makes the feature extraction of SAR images richer and more complete, and has a certain degree of specificity, thereby improving the recognizability and distinguishability of features in SAR images.

[0096] (3) This system uses the classification parameter as a constraint parameter in the process of determining the adaptive processing parameters corresponding to the SAR image to be processed. It can determine the corresponding denoising parameters and enhancement parameters according to the category of each area in the SAR image using the corresponding method, thereby making the determination process of the denoising parameters and enhancement parameters corresponding to each area more targeted.

[0097] (4) This system queries the reference denoising parameters and reference enhancement parameters corresponding to the categories of each area in the SAR image from a pre-established processing parameter database according to the classification parameters, and then corrects the first denoising parameters and first enhancement parameters output by the target deep neural network according to the reference denoising parameters and reference enhancement parameters to obtain the second denoising parameters and second enhancement parameters, which can improve the accuracy and reliability of the denoising parameters and enhancement parameters, thereby improving the denoising and enhancement effects of the SAR image. Example 3

[0098] Please refer to Figure 4 , Figure 4 The figure is a schematic diagram of an application scenario of a SAR image adaptive processing system based on deep learning provided by this embodiment. The application scenario 100 includes: SAR image acquisition device 110, storage device 120, processing device 130, terminal device 140 and network 150. The various components in the application scenario 100 can be connected in a variety of ways. For example, the SAR image acquisition device 110 can be connected to the storage device 120 and / or the processing device 130 through the network 150, or directly connected to the storage device 120 and / or the processing device 130. For another example, the storage device 120 can be directly connected to the processing device 130 or connected through the network 150. For another example, the terminal device 140 can be connected to the storage device 120 and / or the processing device 130 through the network 150, or directly connected to the storage device 120 and / or the processing device 130.

[0099] The SAR image acquisition device 110 can be used to acquire a SAR image, which can be obtained by synthetic aperture radar imaging technology. In some embodiments, the SAR image can be stored in the storage device 120 after being generated by radar acquisition, and the SAR image acquisition device 110 can acquire the SAR image to be processed from the storage device 120. In some embodiments, the SAR image acquisition device 110 can directly acquire images of target objects (such as cities, farmlands, forests, waters, etc.) to obtain the SAR image to be processed.

[0100] In this embodiment, the SAR image acquisition device 110 can send the acquired SAR image to the storage device 120, the processing device 130, the terminal device 140, etc. through the network 150. In some embodiments, the SAR image acquired by the SAR image acquisition device 110 can be processed by the processing device 130. For example, the processing device 130 can determine the corresponding adaptive processing parameters based on the SAR image. In this embodiment, the adaptive processing parameters can be sent to the storage device 120 for recording, or sent to the terminal device 140 for feedback to the user.

[0101] The network 150 may facilitate the exchange of information and / or data. The network 150 may include any suitable network capable of facilitating the exchange of information and / or data of the application scenario 100. In some embodiments, at least one component of the application scenario 100 (e.g., the SAR image acquisition device 110, the storage device 120, the processing device 130, the terminal device 140) may exchange information and / or data with at least one other component in the application scenario 100 through the network 150. For example, the processing device 130 may obtain a SAR image to be processed from the SAR image acquisition device 110 and / or the storage device 120 through the network 150. For another example, the processing device 130 may obtain a user operation instruction from the terminal device 140 through the network 150, and exemplary operation instructions may include but are not limited to calling up a SAR image, reading an adaptive processing parameter determined based on the SAR image, and the like.

[0102] In this embodiment, the network 150 may be any form of wired or wireless network, or any combination thereof. As an example only, the network 150 may include a cable network, a wired network, an optical fiber network, a telecommunications network, an internal network, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, a near field communication (NFC) network, etc., or any combination thereof. In this embodiment, the network 150 may include at least one network access point, and at least one component of the application scenario 100 may be connected to the network 150 through the access point to exchange data and / or information.

[0103] The storage device 120 may store data, instructions and / or any other information. In this embodiment, the storage device 120 may store data obtained from the SAR image acquisition device 110, the processing device 130 and / or the terminal device 140. For example, the storage device 120 may store the SAR image acquired by the SAR image acquisition device 110; for another example, the storage device 120 may store the adaptive processing parameters calculated by the processing device 130. In some embodiments, the storage device 120 may store data and / or instructions used by the processing device 130 to execute or use to complete the exemplary method described in this specification. In this embodiment, the storage device 120 may include a large-capacity memory, a removable memory, a volatile read-write memory, a read-only memory (ROM), etc., or any combination thereof. Exemplary large-capacity memory may include a magnetic disk, an optical disk, a solid-state disk, etc. In this embodiment, the storage device 120 may be implemented on a cloud platform. As an example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

[0104] In this embodiment, the storage device 120 may be connected to the network 150 to communicate with at least one other component in the application scenario 100 (e.g., the SAR image acquisition device 110, the processing device 130, and the terminal device 140). At least one component in the application scenario 100 may access data, instructions, or other information stored in the storage device 120 through the network 150. In some embodiments, the storage device 120 may be directly connected or communicated with one or more components in the application scenario 100 (e.g., the SAR image acquisition device 110, the terminal device 140). In this embodiment, the storage device 120 may be a part of the SAR image acquisition device 110 and / or the processing device 130.

[0105] The processing device 130 may process data and / or information obtained from the SAR image acquisition device 110, the storage device 120, the terminal device 140, and / or other components of the application scenario 100. In this embodiment, the processing device 130 may obtain a SAR image from any one or more of the SAR image acquisition device 110, the storage device 120, or the terminal device 140, and determine corresponding adaptive processing parameters by processing the SAR image. In some embodiments, the processing device 130 may obtain a pre-stored computer instruction from the storage device 120, and execute the computer instruction to implement the SAR image adaptive processing method based on deep learning described in this specification.

[0106] In this embodiment, the processing device 130 may be a single server or a server group. The server group may be centralized or distributed. In this embodiment, the processing device 130 may be local or remote. For example, the processing device 130 may access information and / or data from the SAR image acquisition device 110, the storage device 120, and / or the terminal device 140 through the network 150. For another example, the processing device 130 may be directly connected to the SAR image acquisition device 110, the storage device 120, and / or the terminal device 140 to access information and / or data. In this embodiment, the processing device 130 may be implemented on a cloud platform. For example, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud cloud, a multi-cloud, etc., or any combination thereof.

[0107] The terminal device 140 can receive, send and / or display data. The received data may include data acquired by the SAR image acquisition device 110, data stored in the storage device 120, adaptive processing parameters obtained by the processing device 130, etc. The sent data may include user input data and instructions, etc. For example, the terminal device 140 can send the operation instruction input by the user to the SAR image acquisition device 110 through the network 150 to control the SAR image acquisition device 110 to perform corresponding data acquisition. For another example, the terminal device 140 can send the evaluation instruction input by the user to the processing device 130 through the network 150.

[0108] In this embodiment, the terminal device 140 may include a mobile device 141, a tablet computer 142, a laptop computer 143, etc. or any combination thereof. For example, the mobile device 141 may include a mobile phone, a personal digital assistant (PDA), a dedicated mobile terminal, etc. or any combination thereof. In this embodiment, the terminal device 140 may include an input device (such as a keyboard, a touch screen), an output device (such as a display, a speaker), etc. In this embodiment, the processing device 130 may be a part of the terminal device 140.

[0109] It should be noted that the above description of the application scenario 100 is only for example and explanation, and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the application scenario 100 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification. For example, the SAR image acquisition device 110 may include more or fewer functional components.

[0110] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects that may be produced may be any one or a combination of the above, or may be any other beneficial effects that may be obtained.

[0111] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are suggested in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

[0112] At the same time, this specification uses specific words to describe the embodiments of this specification. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures or characteristics in one or more embodiments of this specification can be appropriately combined.

[0113] In addition, it will be understood by those skilled in the art that various aspects of this specification may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of this specification may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of this specification may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0114] A computer storage medium may include a propagated data signal containing computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination. A computer storage medium may be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, device or apparatus to communicate, propagate or transmit the program for use. The program code on the computer storage medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0115] The computer program codes required for the operation of each part of this specification can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages ​​such as C language, Visual Basic, Fortran2003, Perl, COBOL2002, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy, or other programming languages, etc. The program code can be run entirely on the user's computer, or run on the user's computer as an independent software package, or run partly on the user's computer and partly on a remote computer, or run entirely on a remote computer or processing device. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0116] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in this specification, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some invention embodiments that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing processing device or mobile device.

[0117] Similarly, it should be noted that in order to simplify the description disclosed in this specification and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this specification, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the subject matter of this specification are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0118] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining the digits. Although the numerical domains and parameters used to confirm the breadth of the range in some embodiments of this specification are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

[0119] Each patent, patent application, patent application publication, and other materials, such as articles, books, specifications, publications, documents, etc., cited in this specification is hereby incorporated by reference in its entirety. Except for application history documents that are inconsistent with or conflicting with the contents of this specification, documents that limit the broadest scope of the claims of this specification (currently or later attached to this specification) are also excluded. It should be noted that if the descriptions, definitions, and / or use of terms in the materials attached to this specification are inconsistent or conflicting with the contents described in this specification, the descriptions, definitions, and / or use of terms in this specification shall prevail.

[0120] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A SAR image adaptive processing method based on deep learning, characterized in that: include: Acquire the SAR image to be processed; Extracting multi-scale features from the SAR image to be processed, wherein the multi-scale features include low-frequency features and high-frequency features; The low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; The multi-scale features are input into a trained target deep neural network to obtain adaptive processing parameters corresponding to the SAR image; the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; The SAR image is processed based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

2. The method according to claim 1, characterized in that The step of extracting multi-scale features from the SAR image to be processed comprises: Performing low-pass filtering on the SAR image based on a Gaussian filter to obtain a low-frequency component in the SAR image; Subtracting the low-frequency component from the SAR image to obtain a high-frequency component of the SAR image; Using feature descriptors to extract features of the low-frequency component and the high-frequency component respectively, to obtain low-frequency features and high-frequency features in the SAR image; the feature descriptors include any one or any combination of SIFT, SURF, and HOG; The low-frequency features and the high-frequency features are spliced ​​to obtain multi-scale features of the SAR image.

3. The method according to claim 1 or 2, characterized in that The large-scale structural information includes edges, textures, and shapes, and the small-scale detail information includes scattering points, noise, and interference fringes.

4. The method according to claim 1, characterized in that The adaptive processing parameters corresponding to the SAR image include a classification parameter, a first denoising parameter and a first enhancement parameter; The step of processing the SAR image to be processed based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image includes: According to the classification parameters, querying reference denoising parameters and reference enhancement parameters corresponding to the category to which the SAR image belongs from a pre-established processing parameter database; Perform comprehensive processing according to the reference denoising parameter and the first denoising parameter to obtain a second denoising parameter; Performing comprehensive processing according to the reference enhancement parameter and the first enhancement parameter to obtain a second enhancement parameter; The SAR image is denoised and enhanced based on the second denoising parameter and the second enhancement parameter to obtain a processed SAR image.

5. The method according to claim 4, characterized in that The denoising and enhancing the SAR image based on the second denoising parameter and the second enhancement parameter to obtain a processed SAR image includes: Performing wavelet transform on the SAR image to obtain wavelet coefficients of the SAR image; Performing threshold processing on the wavelet coefficients according to the second denoising parameter to obtain wavelet coefficients after threshold processing; The wavelet coefficients after the threshold processing are subjected to inverse wavelet transform to obtain a denoised SAR image.

6. The method according to claim 5, characterized in that The performing threshold processing on the wavelet coefficients according to the second denoising parameter to obtain the wavelet coefficients after the threshold processing comprises: According to the second denoising parameter, a target threshold function is selected; the target threshold function includes a hard threshold function, a soft threshold function or an improved threshold function; Calculating the absolute value of each component of the wavelet coefficient to obtain the component absolute value; Comparing the component absolute value with the target threshold function to obtain a comparison result; The wavelet coefficients are assigned values ​​according to the comparison result and the target threshold function to obtain processed wavelet coefficients; the assignment process includes keeping unchanged, setting to zero, and subtracting a threshold.

7. The method according to claim 5, characterized in that The denoising and enhancing the SAR image based on the second denoising parameter and the second enhancement parameter to obtain a processed SAR image further includes: Performing histogram statistics on the SAR image to obtain the grayscale distribution of the SAR image; equalize the grayscale distribution according to the second enhancement parameter to obtain a equalized grayscale distribution; The equalized grayscale distribution is inversely mapped to obtain an enhanced SAR image.

8. The method according to claim 7, characterized in that The equalizing the grayscale distribution according to the second enhancement parameter to obtain the equalized grayscale distribution comprises: According to the second enhancement parameter, a target equalization method is selected; the target equalization method includes global equalization, local equalization, and adaptive equalization; According to the target equalization method, the grayscale distribution is transformed to obtain a equalized grayscale distribution.

9. The method according to claim 7, characterized in that The step of performing inverse mapping on the equalized grayscale distribution to obtain an enhanced SAR image comprises: Calculating the cumulative distribution function corresponding to each gray level according to the equalized gray distribution; According to the cumulative distribution function, each gray level is mapped to a new gray level so that the difference between the new gray level and the original gray level is minimized while keeping the equalized gray level distribution unchanged; According to the new grayscale, each pixel value of the SAR image is replaced with a corresponding new grayscale value to obtain an enhanced SAR image.

10. A SAR image adaptive processing system based on deep learning, characterized in that: include: An acquisition module, used for acquiring the SAR image to be processed; A multi-scale feature extraction module, used to extract multi-scale features from the SAR image, wherein the multi-scale features include low-frequency features and high-frequency features, wherein the low-frequency features include large-scale structural information of the SAR image, and the high-frequency features include small-scale detail information of the SAR image; The adaptive processing parameter determination module is used to input the multi-scale features into the trained target deep neural network to obtain the adaptive processing parameters corresponding to the SAR image, wherein the target deep neural network is trained based on the following method: Using multiple groups of sample SAR images as training data to input into an initial deep neural network, wherein each of the sample SAR images is annotated with corresponding sample multi-scale features and sample adaptive processing parameters, wherein the sample adaptive processing parameters include classification parameters, denoising parameters, and enhancement parameters; Using a back propagation algorithm and a gradient descent optimization method, the network parameters of the initial deep neural network are updated so that the distance between the output of the initial deep neural network and the sample adaptive processing parameters corresponding to the training data is less than or equal to a preset threshold, thereby obtaining a target deep neural network; The image processing module is used to process the SAR image based on the adaptive processing parameters corresponding to the SAR image to obtain a processed SAR image.

Images