A SAR image enhancement method based on regional adaptation and related equipment

Abstract

The present application relates to a SAR image enhancement method based on regional adaptation and related equipment. The method includes: performing discrete wavelet transform processing on an initial ocean SAR image to generate initial multi-scale wavelet coefficients; establishing an initial universal hidden Markov tree model and determining a high-noise region image and a low-noise region image; performing discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multi-scale wavelet coefficients and low-noise multi-scale wavelet coefficients; establishing a high-noise universal hidden Markov tree model and a low-noise universal hidden Markov tree model respectively; determining the high-noise region wavelet coefficients and the low-noise region wavelet coefficients; updating the initial universal hidden Markov tree model to obtain a target universal hidden Markov tree model; and performing inverse wavelet transform on the wavelet coefficients in the target universal hidden Markov tree model to obtain a target ocean SAR image.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a SAR image enhancement method based on regional adaptation and related equipment. Background Art

[0002] Marine SAR imagery, acquired using synthetic aperture radar, provides high-resolution images of the ocean surface under various meteorological conditions. However, these images are often subject to noise and lack clear features. Therefore, enhancement techniques are needed to reduce noise, enhance image features, improve visual quality, and increase data processing efficiency, thereby more accurately supporting ocean monitoring and analysis.

[0003] Existing image enhancement technologies include local filtering, non-local filtering, transform domain filtering, and machine / deep learning-based algorithms. Local filtering (such as mean and median filtering) does not adequately preserve complex image details, and non-local filtering is computationally expensive when processing large-scale images. Transform domain filtering (such as frequency domain bandpass filtering) excels in detail preservation but has limited processing capabilities for large-scale images and non-uniform noise. Machine / deep learning methods are not yet ideal for marine SAR image processing. Overall, existing technologies are inefficient and lack sufficient capabilities when processing large-scale and complex background marine SAR images. Summary of the Invention

[0004] The embodiments of the present invention provide a regional adaptive SAR image enhancement method and related equipment, which at least solve the problems of low efficiency and insufficient capability in processing large-scale and complex background ocean SAR images in related technologies.

[0005] According to a first aspect of an embodiment of the present invention, a method for SAR image enhancement based on region adaptation is provided, comprising:

[0006] Performing discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image;

[0007] Based on the initial multi-scale wavelet coefficients, an initial universal hidden Markov tree model corresponding to the initial ocean SAR image is established, and a high-noise region image and a low-noise region image of the initial ocean SAR image are determined, wherein the initial universal hidden Markov tree model is used to determine a hidden state probability set of the initial ocean SAR image;

[0008] Performing discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image;

[0009] Based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively establishing a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image, the high-noise universal hidden Markov tree model being used to determine a hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model being used to determine a hidden state probability set of the low-noise region image;

[0010] Determining the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and determining the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image;

[0011] updating the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients to obtain an updated target universal hidden Markov tree model;

[0012] Performing inverse wavelet transform on the wavelet coefficients in the target universal hidden Markov tree model to obtain an enhanced target ocean SAR image.

[0013] According to a second aspect of an embodiment of the present invention, a SAR image enhancement device based on regional adaptation is provided, comprising:

[0014] A generating module, configured to perform discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image;

[0015] a processing module, configured to establish an initial universal hidden Markov tree model corresponding to the initial ocean SAR image based on the initial multi-scale wavelet coefficients, and determine a high-noise region image and a low-noise region image of the initial ocean SAR image, wherein the initial universal hidden Markov tree model is used to determine a hidden state probability set of the initial ocean SAR image;

[0016] The generating module is further configured to perform discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image;

[0017] An establishment module is used to establish a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively, wherein the high-noise universal hidden Markov tree model is used to determine the hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model is used to determine the hidden state probability set of the low-noise region image;

[0018] a determination module, configured to determine the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and to determine the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image;

[0019] an updating module, configured to update the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients, to obtain an updated target universal hidden Markov tree model;

[0020] The transformation module is used to perform inverse wavelet transformation on the wavelet coefficients in the target universal hidden Markov tree model to obtain the target ocean SAR image after image enhancement.

[0021] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a processor, and a memory storing a program, wherein the program comprises instructions that, when executed by the processor, cause the processor to execute the method according to the first aspect.

[0022] According to a fourth aspect of an embodiment of the present invention, a non-transitory machine-readable medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method according to the first aspect.

[0023] The region-adaptive SAR image enhancement method provided by an embodiment of the present invention can improve the efficiency and effectiveness of processing large-scale and complex-background ocean SAR images. First, a discrete wavelet transform is performed on the initial ocean SAR image to generate initial multiscale wavelet coefficients. Then, based on these coefficients, an initial universal hidden Markov tree (uHMT) model is established to identify high-noise and low-noise regions in the image, and a universal hidden Markov tree model is established for each region. Further wavelet transforms are performed on these regions to obtain multiscale wavelet coefficients for the high-noise and low-noise regions. Using these coefficients, the coefficients corresponding to these regions in the initial uHMT model are updated, and enhanced wavelet coefficients for the high-noise and low-noise regions are calculated. The updated uHMT model is then subjected to an inverse wavelet transform to obtain an enhanced target SAR image. This method can adaptively adjust the processing strategy and optimize the enhancement effect for regions with different noise levels, thereby improving image clarity and detail retention. It is particularly suitable for processing large-scale and complex-background ocean SAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be derived from these drawings without inventive effort.

[0025] Figure 1 A schematic flow chart of a SAR image enhancement method based on regional adaptation is provided in an embodiment of the present invention.

[0026] Figure 2 A schematic diagram of an initial ocean SAR image provided by an embodiment of the present invention.

[0027] Figure 3 A schematic diagram of a target ocean SAR image provided by an embodiment of the present invention.

[0028] Figure 4 A schematic structural diagram of a SAR image enhancement device based on regional adaptation provided by an embodiment of the present invention.

[0029] Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following describes embodiments of the present invention in more detail with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0031] Marine synthetic aperture radar (SAR) imagery, acquired through synthetic aperture radar (SAR), can provide high-resolution ocean surface images under various meteorological conditions, making it a valuable tool for ocean monitoring and analysis. However, despite these advantages, SAR images are often subject to noise, and their detailed features are often unclear. These issues complicate image interpretation and analysis, impacting the accuracy of marine environmental monitoring. Therefore, image enhancement technology plays a crucial role in marine SAR image processing. Its goal is to reduce noise, enhance image details, improve visual quality, and enhance data processing efficiency to more accurately support ocean monitoring and analysis tasks.

[0032] Currently, commonly used image enhancement techniques include local filtering, non-local filtering, transform domain filtering, and algorithms based on machine learning and deep learning. However, each of these techniques has its limitations. Although local filtering methods (such as mean filtering and median filtering) can reduce noise to a certain extent, they perform poorly in preserving complex image details and often result in blurred images. Non-local filtering methods can better preserve image details, but their computational cost is extremely high when processing large-scale images, limiting their effectiveness in practical applications. Transform domain filtering (such as frequency-domain bandpass filtering) performs well in preserving details, but its ability to process large-scale images and non-uniform noise is limited, which limits its application in various complex marine environments. Although machine learning and deep learning-based methods have shown great potential in the field of image processing, their effectiveness in marine SAR image enhancement is still unstable and requires a large amount of annotated data and computing resources.

[0033] Overall, existing image enhancement technologies are inefficient and limited in their capabilities when used for large-scale, complex background ocean SAR imagery. These shortcomings not only affect the clarity and detail of SAR images, but also limit the accuracy and reliability of ocean monitoring and analysis. Therefore, developing more efficient and accurate ocean SAR image enhancement technologies to provide high-quality image data in various complex environments has become an urgent challenge.

[0034] In order to solve the above problems, an embodiment of the present invention provides a SAR image enhancement method based on regional adaptation and related devices.

[0035] Figure 1 The following is a flow chart of a SAR image enhancement method based on regional adaptation provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps.

[0036] Step S101 : performing discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image.

[0037] Step S102: Based on the initial multi-scale wavelet coefficients, an initial general hidden Markov tree model corresponding to the initial ocean SAR image is established, and a high-noise area image and a low-noise area image of the initial ocean SAR image are determined. The initial general hidden Markov tree model is used to determine the implicit state probability set of the initial ocean SAR image.

[0038] Step S103 , performing discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image.

[0039] Step S104: Based on the high-noise multi-scale wavelet coefficients and the low-noise multi-scale wavelet coefficients, a high-noise universal hidden Markov tree model corresponding to the high-noise area image and a low-noise universal hidden Markov tree model corresponding to the low-noise area image are respectively established. The high-noise universal hidden Markov tree model is used to determine the implicit state probability set of the high-noise area image, and the low-noise universal hidden Markov tree model is used to determine the implicit state probability set of the low-noise area image.

[0040] Step S105 , determining the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and determining the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image.

[0041] Step S106 , based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients, the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model are updated to obtain an updated target universal hidden Markov tree model.

[0042] Step S107 , performing inverse wavelet transform on the wavelet coefficients in the target general hidden Markov tree model to obtain an enhanced target ocean SAR image.

[0043] Firstly, the initial ocean SAR image is processed by discrete wavelet transform to generate the initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image.

[0044] In this embodiment, the initial ocean SAR image is first processed using a discrete wavelet transform. Discrete wavelet transform is a technique that captures image information in the spatial and frequency domains by decomposing the signal's different frequency components. By combining high-pass and low-pass filters, the initial ocean SAR image is decomposed into high- and low-frequency components of varying scales. This decomposition effectively captures edge and texture features in the initial ocean SAR image.

[0045] Specifically, high-frequency components are used to represent image details, such as edges and textures. These high-frequency components are often more susceptible to noise. Low-frequency components are used to represent global information and relatively smooth areas of the image.

[0046] In this embodiment, by processing the initial ocean SAR image, the generated initial multi-scale wavelet coefficients contain the above-mentioned high-frequency and low-frequency features, and are used in the subsequent noise detection and processing process.

[0047] After generating initial multiscale wavelet coefficients corresponding to the initial ocean SAR image, an initial general hidden Markov tree model corresponding to the initial ocean SAR image is established based on the initial multiscale wavelet coefficients, and images of high-noise regions and low-noise regions of the initial ocean SAR image are determined. In this embodiment, the initial general hidden Markov tree model is used to determine a hidden state probability set for the initial ocean SAR image.

[0048] In this embodiment, an initial universal hidden Markov tree model can be established based on the initial multi-scale wavelet coefficients generated in the above steps. This initial universal hidden Markov tree model is a statistical model that models data dependencies at different scales. It can effectively process the multi-scale features of images and can capture the dependencies between different scales.

[0049] Specifically, the general hidden Markov tree model can treat the wavelet coefficients in the initial ocean SAR image as a set of random variables and describe the dependencies between these variables through the state transition relationship between parent and child nodes. It can capture the probabilistic relationship between multiple scales.

[0050] Furthermore, by analyzing the statistical properties of the initial multi-scale wavelet coefficients, the initial general hidden Markov tree model can separate high-noise and low-noise regions in the image. In practical applications, this can be distinguished based on the amplitude and scale of the wavelet coefficients: regions with higher noise have higher amplitudes.

[0051] Afterwards, discrete wavelet transform is performed on the high-noise region image and the low-noise region image respectively to generate high-noise multi-scale wavelet coefficients corresponding to the high-noise region image and low-noise multi-scale wavelet coefficients corresponding to the low-noise region image.

[0052] In this embodiment, the high-noise and low-noise region images identified in the above steps are each subjected to a new discrete wavelet transform. Images in these high-noise regions may contain more noise components, and the discrete wavelet transform captures the high-frequency wavelet coefficients in these regions, forming high-noise multiscale wavelet coefficients. Images in low-noise regions are relatively stable and contain less noise, so the discrete wavelet transform captures relatively smooth wavelet coefficients, forming low-noise multiscale wavelet coefficients.

[0053] After obtaining the high-noise multiscale wavelet coefficients corresponding to the high-noise region image and the low-noise multiscale wavelet coefficients corresponding to the low-noise region image, a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image can be established based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively. In this embodiment, the high-noise universal hidden Markov tree model is used to determine the hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model is used to determine the hidden state probability set of the low-noise region image.

[0054] In this embodiment, based on the initial universal hidden Markov tree model, hidden Markov tree models are separately constructed for different noise feature areas.

[0055] Specifically, a high-noise universal hidden Markov tree model is built based on high-noise multiscale wavelet coefficients to describe the statistical characteristics of high-noise regions and capture the probabilistic relationships between wavelet coefficients within these regions. A low-noise universal hidden Markov tree model is built based on low-noise multiscale wavelet coefficients to describe the relatively stable statistical characteristics of low-noise regions. By establishing separate high-noise and low-noise models, subsequent image processing can be performed more accurately.

[0056] Afterwards, the wavelet coefficients of the high-noise region are determined according to the implicit state probability set of the high-noise region image, and the wavelet coefficients of the low-noise region are determined according to the implicit state probability set of the low-noise region image.

[0057] In this embodiment, new wavelet coefficients for high-noise and low-noise regions are estimated by utilizing the implicit state probability set calculated within the hidden Markov tree model. The implicit state probability set, derived through model training, represents the probability distribution of wavelet coefficients under varying noise conditions. Based on this implicit state probability set, wavelet coefficients closer to their true values ​​can be estimated, eliminating the effects of noise on the coefficients.

[0058] The initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model are updated based on the wavelet coefficients in the high-noise region and the wavelet coefficients in the low-noise region to obtain an updated target universal hidden Markov tree model.

[0059] In this embodiment, the updated wavelet coefficients estimated from high-noise and low-noise regions are used to adjust the initial hidden Markov tree model. Through this feedback update process, the model's accuracy is improved by replacing the original coefficients with more accurate ones. Through these steps, the initial general hidden Markov tree model becomes more accurate, reflecting the true characteristics of different regions in the current SAR image.

[0060] Finally, the wavelet coefficients in the target general hidden Markov tree model are subjected to inverse wavelet transform to obtain the enhanced target ocean SAR image.

[0061] In this embodiment, the wavelet coefficients in the updated and optimized hidden Markov tree model are subjected to an inverse wavelet transform. The multi-scale wavelet coefficients are reassembled into an image, thereby generating the final enhanced ocean SAR image. Specifically, the inverse wavelet transform is a process that reconstructs the image data after wavelet decomposition back into the original image space.

[0062] Based on this method, the multi-scale properties of the discrete wavelet transform and hidden Markov tree model can be effectively utilized to enhance SAR images. From initial image decomposition to modeling of regional noise characteristics, and then to updating and inverse transformation, image enhancement and noise suppression are ultimately achieved. This method is suitable for processing noisy SAR images, making target information within the image easier to analyze and identify.

[0063] In an optional embodiment, the initial multi-scale wavelet coefficients include a first high-frequency characteristic wavelet coefficient and a first low-frequency characteristic wavelet coefficient, the high-noise multi-scale wavelet coefficients include a second high-frequency characteristic wavelet coefficient and a second low-frequency characteristic wavelet coefficient, and the low-noise multi-scale wavelet coefficients include a third high-frequency characteristic wavelet coefficient and a third low-frequency characteristic wavelet coefficient.

[0064] In this embodiment, when performing discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image, the initial ocean SAR image can be processed respectively by a high-pass filter and a low-pass filter based on the Haar wavelet basis function to generate a first high-frequency characteristic wavelet coefficient and a first low-frequency characteristic wavelet coefficient.

[0065] The Haar wavelet basis function defines a specific filter structure that decomposes and reconstructs images using a pair of mother wavelets and a scaling function. In SAR image processing applications, the Haar wavelet, due to its simple computational characteristics, can quickly decompose images into multiple scales, making it suitable for real-time or near-real-time image enhancement.

[0066] In the wavelet transform process, the image is decomposed through a set of filters. For Haar wavelets, these filters include a high-pass filter and a low-pass filter. The application of these filters generates different frequency components of the image, allowing the image's features to be analyzed at different scales:

[0067] Low-pass filter: This filter is used to capture low-frequency information (i.e., smooth areas or general structures) in an image, generating low-frequency characteristic wavelet coefficients. Low-frequency information typically preserves the primary structure of the image and has a higher energy density. The output of the low-pass filter represents a coarse version of the image, capturing large-scale image features.

[0068] High-pass filter: This filter extracts high-frequency information (i.e., edges and details) from an image and generates high-frequency feature wavelet coefficients. High-frequency information typically includes image details such as texture, edges, and noise. The output of the high-pass filter provides detailed image information that can be used for further feature analysis.

[0069] After applying the high-pass filter and the low-pass filter, a first high-frequency characteristic wavelet coefficient and a first low-frequency characteristic wavelet coefficient are obtained.

[0070] The first high-frequency feature wavelet coefficient is used to represent the detailed features of the image, especially edge and texture information. It is captured through a high-pass filter and is significant in high-noise areas or edge features of the image, and can be used to enhance image detail and sharpness. The first low-frequency feature wavelet coefficient is used to represent the low-frequency information of the image, representing the basic structure and smooth parts of the image. The low-frequency coefficient is obtained through a low-pass filter, retaining the main energy and overall shape of the image. Low-frequency information is mainly used to maintain the basic shape of the image and reduce noise.

[0071] After performing the wavelet decomposition, the first high-frequency characteristic wavelet coefficients and the first low-frequency characteristic wavelet coefficients are stored for use in subsequent steps. These coefficients can be stored in the form of a matrix or multidimensional array to facilitate further processing and calculation. For example, in a hidden Markov tree model, these coefficients will be used to build the initial model and further determine the hidden state probability set of the image.

[0072] The initial ocean SAR image is processed using a discrete wavelet transform using Haar wavelet basis functions to capture the different frequency characteristics of the image. The application of low-pass and high-pass filters effectively decomposes the image at different scales, thereby extracting the high-frequency and low-frequency characteristics of the image.

[0073] In an optional embodiment, an initial general hidden Markov tree model corresponding to the initial ocean SAR image may be established based on the initial multi-scale wavelet coefficients.

[0074] Specifically, the first probability density function of the Gaussian mixture model of each initial multiscale wavelet coefficient can be determined, and the first node relationship between each initial multiscale wavelet coefficient at different scale layers can be determined using the parameter set of the hidden Markov tree model. Then, an initial universal hidden Markov tree model is established based on the first probability density function and the first node relationship.

[0075] In this embodiment, a mixed Gaussian model can first be constructed using the initial multi-scale wavelet coefficients. The mixed Gaussian model is a probabilistic model that assumes that a data point is composed of multiple Gaussian distributions (usually referred to as components). Each of these Gaussian components has its own mean, variance, and weight. The following is a specific implementation method for this step:

[0076] Acquisition of multi-scale wavelet coefficients: The initial ocean SAR image is processed by discrete wavelet transform to generate initial multi-scale wavelet coefficients. These initial multi-scale wavelet coefficients reflect the characteristic information of the image at different scales.

[0077] Establish a Gaussian mixture model: Treat each initial multi-scale wavelet coefficient as a data point and assume that these data points can be represented by a Gaussian mixture model. Use the expectation maximization (EM) algorithm or other suitable estimation algorithm to fit a Gaussian model with K components.

[0078] The first probability density function of the Gaussian mixture model can be determined based on the following formula (1):

[0079]

[0080] Where x is the initial multi-scale wavelet coefficient; p(x) is the first probability density function of x; K is the number of Gaussian distribution components in the mixed Gaussian model; π k is the prior probability of state k, N(x,μ k ,∑ k ) is the probability of x given the kth Gaussian distribution component, μ k is the mean vector of the kth Gaussian distribution, ∑ k is the covariance matrix of the k-th Gaussian distribution.

[0081] At the same time, a Hidden Markov Tree (HMT) model is used to capture the dependencies between wavelet coefficients at different scales. The HMT is a graphical model that can effectively describe probabilistic dependencies in hierarchical structures and is particularly suitable for multi-scale analysis, such as wavelet transforms.

[0082] Each node in a hidden Markov tree model corresponds to a wavelet coefficient, and the connections between nodes represent their dependencies at different scales. The root node corresponds to the coefficients at the lowest scale, and the leaf nodes correspond to the coefficients at the highest scale. Each parent node has a state transition probability between its child nodes.

[0083] In this embodiment, the parameter set of the hidden Markov tree model can be determined based on the following formula (2):

[0084]

[0085] Among them, θ is the parameter set of the hidden Markov tree model; p s1 (m) is the state probability distribution function of the root node state s1, which represents the probability of the root node being in the implicit state m; is the state transition probability between parent node i and child node ρ(i), which means the probability that the child node transitions to state n when the parent node is in state m; μ i,m is the mean of the probability density function of the wavelet coefficients of the current node i in the implicit state m; is the variance of the probability density function of the wavelet coefficient of the current node i in the implicit state m.

[0086] The parameter set of the hidden Markov tree model can be used to define the node relationships between the initial multi-scale wavelet coefficients at different scale levels, including the state transition probabilities between parent and child nodes at different levels and the statistical characteristics of the corresponding wavelet coefficients.

[0087] The implicit state probability set of the initial ocean SAR image is determined based on the following method: based on the meta-parameters of the initial general hidden Markov tree model, the implicit state probability set is determined by the maximum expectation algorithm.

[0088] The hidden state probability set is one of the core results output by the hidden Markov tree model, which can be used to represent the probability of the signal and noise in different states under each wavelet coefficient.

[0089] In practical applications, the EM algorithm can be used to calculate the implicit state probability of each wavelet coefficient in the initial ocean SAR image based on an initial general hidden Markov tree model. The algorithm is iteratively updated until it converges to a stable state.

[0090] In an optional embodiment, the high-noise region image and the low-noise region image of the initial ocean SAR image may be determined based on the initial multi-scale wavelet coefficients.

[0091] In this embodiment, sliding window processing can be performed on the initial ocean SAR image based on a window of a preset size to determine the median of the first scale HH subband of multiple initial multi-scale wavelet coefficients in each window; then the median corresponding to each window is sorted, and the sorting result is output to determine the high-noise area image and the low-noise area image based on the sorting result.

[0092] In this embodiment, the sliding window processing can analyze the noise characteristics of the image in a local area, thereby effectively distinguishing high-noise and low-noise areas. By dividing the image into multiple small windows, the noise distribution can be observed at a finer scale.

[0093] Specifically, a fixed-size window (e.g., 64x64, 128x128, or 256x256 pixels) can be defined and gradually slid across the initial ocean SAR image from left to right and from top to bottom. Each time the window slides, it covers a local area of ​​the image. During the sliding process, the window position gradually moves, one or more pixels at a time (as needed), until the entire image is covered.

[0094] Through the sliding window operation, the initial image is divided into multiple local regions, each of which is covered by a window. Each window contains several pixel values ​​of the image for subsequent noise analysis.

[0095] In this embodiment, the wavelet coefficients of the first scale HH subband are mainly used to capture high-frequency details in the image, which are usually greatly affected by noise. Therefore, extracting the median of the HH subband can help identify the noise intensity.

[0096] Specifically, a multiscale wavelet transform is performed on the initial ocean SAR image to obtain wavelet coefficients for the first-scale HH subband. These coefficients reflect the intensity of detail and edge features in the image. For each sliding window region, the wavelet coefficients corresponding to the first-scale HH subband are extracted for all pixels within that region. The median of these wavelet coefficients is then calculated. Median calculation is a robust statistical method that is less susceptible to noise interference and effectively reflects the typical wavelet coefficient values ​​within the region.

[0097] Each sliding window area corresponds to a median value, which represents the central tendency of high-frequency features in the local area and is used to distinguish noise characteristics.

[0098] In this embodiment, the sorted median value can be used to compare the noise levels of various local areas, thereby determining which areas have higher noise characteristics.

[0099] Specifically, the median values ​​of all sliding window areas are collected into a list or array. These medians are then sorted from lowest to highest, or according to a specific sorting rule (such as descending or ascending). Sorting can use common sorting algorithms such as quick sort or merge sort.

[0100] The sorted median sequence clearly shows the distribution of noise intensity across different regions of the image. The termination sequence is arranged from largest to smallest, with the first values ​​representing areas with lower noise levels, and the last values ​​representing areas with higher noise levels. In practical applications, the first 5% of median values ​​can be considered high-noise areas, while the last 5% can be considered low-noise areas.

[0101] In an optional embodiment, when establishing a high-noise universal hidden Markov tree model corresponding to a high-noise region image and a low-noise universal hidden Markov tree model corresponding to a low-noise region image based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively, the following steps may be performed:

[0102] First, the second probability density functions of the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients are determined. Then, the second node relationships between the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients at different scale levels are determined using the parameter set of the hidden Markov tree model. Finally, based on the second probability density functions and the second node relationships, a high-noise universal hidden Markov tree model and a low-noise universal hidden Markov tree model are established, respectively.

[0103] In this embodiment, during the construction of the universal hidden Markov tree model, the probability density function is used to describe the statistical characteristics of the wavelet coefficients. This step can define statistical models for the wavelet coefficients in the high noise region and the low noise region respectively.

[0104] In high-noise regions, where the signal is significantly affected by noise, the distribution characteristics of the wavelet coefficients may differ from those in low-noise regions. By determining the probability density function (PDF) in high-noise regions, we can effectively describe the statistical characteristics of these coefficients. Specifically, a Gaussian mixture model can be used for this purpose, as it can flexibly adapt to the varying distribution characteristics of data, particularly in complex natural scenes.

[0105] Multi-scale wavelet coefficients in low-noise regions generally exhibit more stable and concentrated distributions. Statistical modeling of these coefficients can help accurately distinguish between signal and noise during image enhancement.

[0106] In this embodiment, the hidden Markov tree model captures the dependencies between multi-scale wavelet coefficients through the relationships between nodes. The state transition relationships between nodes can be established using wavelet coefficients at different scale levels.

[0107] In a hidden Markov tree structure, each wavelet coefficient can be considered a node. The relationships between parent and child nodes reflect the information transfer between different scales. For example, a parent node at a coarse scale may control the state distribution of several child nodes at finer scales. By defining these node relationships, the model can capture the signal and noise characteristics at multiple scales.

[0108] The state transition relationship between nodes is typically represented by a transition probability matrix. Each element of the matrix describes the probability of a child node changing state when the parent node is in a certain state. Different hidden Markov tree models have different parameter sets to represent node relationships under different noise characteristics.

[0109] In this embodiment, a high-noise and low-noise universal hidden Markov tree model may be established based on the second probability density function and the second node relationship.

[0110] Specifically, based on the second probability density function and the second node relationship, a hidden Markov tree model specifically for high-noise regions is constructed. This model can describe the distribution and inter-scale relationships of multi-scale wavelet coefficients in high-noise regions, helping to effectively distinguish and remove noise during the enhanced processing process.

[0111] Similarly, based on the second probability density function and the second node relationship in the low-noise region, a hidden Markov tree model suitable for low-noise regions can be constructed. Because the signal in low-noise regions is more stable, this model can more accurately preserve the signal's detailed features while reducing the impact of noise.

[0112] The statistical characteristics and node relationships in the high-noise hidden Markov tree model are used to reduce the impact of noise and retain or enhance useful signal information. The low-noise hidden Markov tree model is used to accurately preserve the low-noise detail characteristics of the image and avoid detail loss caused by over-smoothing.

[0113] In an optional embodiment, when determining the wavelet coefficients of the high-noise region based on the implicit state probability set of the high-noise region image, and determining the wavelet coefficients of the low-noise region based on the implicit state probability set of the low-noise region image, the conditional means corresponding to the high-noise region image and the low-noise region image can be determined based on the following formula (3):

[0114]

[0115] Among them, Y i Represents the wavelet coefficients of the signal in the image; W i Represents the wavelet coefficients of the parent node at different scales; S i represents the implicit state of node i; m represents the specific value of the implicit state; represents the variance of the wavelet coefficient of node i in the hidden state m; σ 2 represents the variance of the noise; ω i Represents the wavelet coefficient of the parent node of node i;

[0116] Based on the implicit state probability set of the high-noise region image and the implicit state probability set of the low-noise region image, the wavelet coefficients of the high-noise region and the wavelet coefficients of the low-noise region are determined based on the following formula (4):

[0117]

[0118] in, Represents the new wavelet coefficients, y i is the wavelet coefficient representing the signal in the local image (high noise area image or low noise area image).

[0119] The SAR image enhancement method based on regional adaptation provided by the embodiment of the present invention is described below with reference to specific embodiments.

[0120] In this embodiment, it is assumed that Figure 2 is the initial ocean SAR image, and image enhancement processing is performed on it.

[0121] First, a discrete wavelet transform (DWT) is performed on the initial ocean SAR image. Specifically, a set of high-pass and low-pass filters are applied to decompose the image into high- and low-frequency components of varying scales. High-frequency components primarily represent image details, such as edges and textures, which are often significantly affected by noise. Low-frequency components represent the overall structure and smoother areas of the image, which contain less noise.

[0122] Assuming the initial ocean SAR image size is 512x512 pixels, the image is decomposed into one low-frequency sub-band and three high-frequency sub-bands (horizontal, vertical, and diagonal) through discrete wavelet transform. These sub-bands form the initial set of multi-scale wavelet coefficients, recording the frequency characteristics of the image at different scales.

[0123] Based on the initial multi-scale wavelet coefficients, an initial universal hidden Markov tree (uHMT) model is established. The hidden Markov tree model is a statistical model used to capture the dependencies between multiple scales.

[0124] Each coefficient in the initial multi-scale wavelet coefficients is regarded as a node in the hidden Markov tree model.

[0125] The connections between nodes represent the dependency of wavelet coefficients at different scales, where the parent node corresponds to the low-frequency coefficients and the child node corresponds to the high-frequency coefficients.

[0126] In this example, we assume that the initial HMT model may contain a three-layer structure, with each layer representing a different scale. Parent nodes represent high-scale features (low-frequency information), while child nodes represent low-scale features (high-frequency information). By training this model, we can obtain the state probability distribution of each wavelet coefficient in the image, providing a basis for subsequent noise separation and feature enhancement.

[0127] The initial universal hidden Markov tree model can be used to determine the high-noise region image and the low-noise region image. Then, these regions are processed again with discrete wavelet transform to generate multi-scale wavelet coefficients for the high-noise and low-noise regions.

[0128] In high-noise areas, the amplitude of the wavelet coefficients is higher, which usually represents more details and texture features. In low-noise areas, the amplitude of the wavelet coefficients is lower, which represents smoother and more stable image areas.

[0129] For example, for images in high-noise areas, discrete wavelet transforms may generate more prominent high-frequency features, which are then used for noise filtering and image sharpening in subsequent steps. Low-noise areas retain their original smoothness, helping to maintain the overall stability of the image.

[0130] Based on the wavelet coefficients of high-noise and low-noise regions, universal hidden Markov tree models are established for high-noise and low-noise regions, respectively. These models are used to capture the probabilistic dependencies between regions with different noise characteristics.

[0131] The high-noise HMT model describes the statistical characteristics of high-noise regions, including the state transition probability between wavelet coefficients. The low-noise HMT model is used to capture the relatively stable characteristics of low-noise regions.

[0132] By analyzing the high-frequency wavelet coefficients in high-noise regions, a high-noise HMT model is established that can effectively identify and filter out noise in these regions while preserving image details and edges. Similarly, a low-noise HMT model ensures that smooth and continuous image structures are maintained within low-noise regions.

[0133] Based on the implicit state probability sets of the high-noise and low-noise area images, the wavelet coefficients of the high-noise area and the low-noise area are estimated respectively. In this way, the denoised wavelet coefficients can be obtained, thereby reducing the impact of noise on image details.

[0134] The implicit state probability set represents the probability distribution of the wavelet coefficients under different noise conditions, and the initial wavelet coefficients are adjusted through these probability distributions.

[0135] In high-noise regions, the latent state probabilities calculated using the high-noise HMT model are used to adjust the high-frequency wavelet coefficients, making them closer to the actual image features and reducing the errors introduced by noise. Similarly, the wavelet coefficients in low-noise regions are adjusted using the low-noise HMT model to preserve the smoothness of the image.

[0136] The estimated wavelet coefficients of the high-noise and low-noise regions are used to update the initial universal hidden Markov tree model, thereby obtaining a more accurate target universal hidden Markov tree model. The update process replaces the initial coefficients with the estimated wavelet coefficients to make the model more consistent with the characteristics of the actual image.

[0137] Assuming that the high-frequency coefficients of a node in the original model are greatly affected by noise, the updated wavelet coefficients can more accurately reflect the edge information in the actual image, thereby improving the model's ability to describe image features.

[0138] Finally, an inverse wavelet transform is performed on the wavelet coefficients in the target general hidden Markov tree model to generate an enhanced target ocean SAR image. The inverse wavelet transform process combines the multi-scale wavelet coefficients back into the original image space to reconstruct an image with noise suppression and feature enhancement.

[0139] Figure 3 The target ocean SAR image after image enhancement processing provided by the embodiment of the present invention. Figure 3 As shown in Figure 1, the inverse wavelet transform restores the information in the model to an enhanced image, eliminating noise and improving image detail clarity. Compared to the original image, the image obtained after the inverse transform has significantly less noise, sharper edges, and the target object is more clearly visible in the image.

[0140] Based on the regional adaptive SAR image enhancement method provided by the embodiment of the present invention, the embodiment of the present invention also provides a regional adaptive marine synthetic aperture radar image enhancement device. Figure 4 As shown, the apparatus includes a generating module 401 , a processing module 402 , a establishing module 403 , a determining module 404 , an updating module 405 and a transforming module 406 .

[0141] A generating module 401 is configured to perform discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image;

[0142] Processing module 402 is used to establish an initial universal hidden Markov tree model corresponding to the initial ocean SAR image based on the initial multi-scale wavelet coefficients and determine the high-noise region image and the low-noise region image of the initial ocean SAR image, wherein the initial universal hidden Markov tree model is used to determine the implicit state probability set of the initial ocean SAR image;

[0143] The generating module 401 is further configured to perform discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image;

[0144] Establishing module 403, for establishing a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively, wherein the high-noise universal hidden Markov tree model is used to determine the hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model is used to determine the hidden state probability set of the low-noise region image;

[0145] A determination module 404 is configured to determine the wavelet coefficients of the high-noise region based on the implicit state probability set of the high-noise region image, and to determine the wavelet coefficients of the low-noise region based on the implicit state probability set of the low-noise region image;

[0146] An updating module 405 is configured to update the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients to obtain an updated target universal hidden Markov tree model;

[0147] The transformation module 406 is used to perform inverse wavelet transformation on the wavelet coefficients in the target general hidden Markov tree model to obtain an enhanced target ocean SAR image.

[0148] In this embodiment, the initial multi-scale wavelet coefficients include a first high-frequency characteristic wavelet coefficient and a first low-frequency characteristic wavelet coefficient, the high-noise multi-scale wavelet coefficients include a second high-frequency characteristic wavelet coefficient and a second low-frequency characteristic wavelet coefficient, and the low-noise multi-scale wavelet coefficients include a third high-frequency characteristic wavelet coefficient and a third low-frequency characteristic wavelet coefficient.

[0149] In this embodiment, the generating module 401 is specifically configured to process the initial ocean SAR image using a high-pass filter and a low-pass filter based on the Haar wavelet basis function to generate first high-frequency characteristic wavelet coefficients and first low-frequency characteristic wavelet coefficients.

[0150] In this embodiment, the processing module 402 is specifically used to determine a first probability density function of a mixture Gaussian model of each initial multi-scale wavelet coefficient, and to determine a first node relationship between different scale layers of each initial multi-scale wavelet coefficient through a parameter set of a hidden Markov tree model.

[0151] establishing an initial universal hidden Markov tree model based on the first probability density function and the first node relationship;

[0152] The first probability density function is determined based on the following formula (1):

[0153]

[0154] Where x is the initial multi-scale wavelet coefficient; p(x) is the first probability density function of x; K is the number of Gaussian distribution components in the mixed Gaussian model; π k is the prior probability of state k, N(x,μ k ,Σ k ) is the probability of x given the kth Gaussian distribution component, μ k is the mean vector of the kth Gaussian distribution, Σ k is the covariance matrix of the k-th Gaussian distribution;

[0155] The parameter set of the hidden Markov tree model is determined based on the following formula (2):

[0156]

[0157] Among them, θ is the parameter set of the hidden Markov tree model; p s1 (m) is the state probability distribution function of the root node state s1, which represents the probability of the root node being in the implicit state m; is the state transition probability between parent node i and child node ρ(i), which means the probability that the child node transitions to state n when the parent node is in state m; μ i,m is the mean of the probability density function of the wavelet coefficients of the current node i in the implicit state m; is the variance of the probability density function of the wavelet coefficient of the current node i in the implicit state m;

[0158] The implicit state probability set of the initial ocean SAR image is determined based on the following method:

[0159] Based on the meta-parameters of the initial universal hidden Markov tree model, the hidden state probability set is determined by the maximum expectation algorithm.

[0160] In this embodiment, the processing module 402 is specifically used to perform sliding window processing on the initial ocean SAR image based on a window of a preset size, determine the median of the first scale HH subband of multiple initial multi-scale wavelet coefficients in each window; sort the medians corresponding to each window, output the sorting results, and determine the high-noise area image and the low-noise area image based on the sorting results.

[0161] In this embodiment, module 403 is established, which is specifically used to determine the second probability density function of the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients; determine the second node relationship between the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients at different scale layers through the parameter set of the hidden Markov tree model; based on the second probability density function and the second node relationship, a high-noise universal hidden Markov tree model and a low-noise universal hidden Markov tree model are established respectively.

[0162] In this embodiment, the conditional means corresponding to the high-noise region image and the low-noise region image are determined based on the following formula (3):

[0163]

[0164] Among them, Y i Represents the wavelet coefficients of the signal in the image; W i Represents the wavelet coefficients of the parent node at different scales; S i represents the implicit state of node i; m represents the specific value of the implicit state; represents the variance of the wavelet coefficient of node i in the hidden state m; σ 2 represents the variance of the noise; ω i Represents the wavelet coefficient of the parent node of node i;

[0165] Based on the implicit state probability set of the high-noise region image and the implicit state probability set of the low-noise region image, the wavelet coefficients of the high-noise region and the wavelet coefficients of the low-noise region are determined based on the following formula (4):

[0166]

[0167] in, Represents the new wavelet coefficients, y i is the wavelet coefficient representing the signal in the local image (high noise area image or low noise area image).

[0168] An embodiment of the present invention further provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, wherein the computer program, when executed by the at least one processor, causes the electronic device to perform the method of an embodiment of the present invention.

[0169] An embodiment of the present invention further provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to perform the method of the embodiment of the present invention.

[0170] An embodiment of the present invention further provides a computer program product, including a computer program, wherein when the computer program is executed by a processor of a computer, it is used to enable the computer to perform the method of the embodiment of the present invention.

[0171] refer to Figure 5 , a block diagram of an electronic device that can serve as a server or client of an embodiment of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic devices are intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0172] like Figure 5 As shown, the electronic device includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. Various programs and data required for the operation of the electronic device can also be stored in the RAM 503. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0173] Multiple components within the electronic device are connected to the I / O interface 505, including an input unit 506, an output unit 507, a storage unit 508, and a communication unit 509. The input unit 506 can be any type of device capable of inputting information into the electronic device. The input unit 506 can receive input numeric or character information and generate key signal inputs related to user settings and / or function control of the electronic device. The output unit 507 can be any type of device capable of presenting information and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 508 can include, but is not limited to, a magnetic disk or an optical disk. The communication unit 509 allows the electronic device to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0174] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention may be implemented as a computer program that is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be stored in a ROM.

[0175] 502 and / or the communication unit 509 are loaded and / or installed on the electronic device. In some embodiments, the computing unit 501 can be configured to execute the above method in any other appropriate manner (for example, by means of firmware).

[0176] The computer programs for implementing the methods of the embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer programs are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0177] In the context of an embodiment of the present invention, a machine-readable medium can be a tangible medium that can contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] It should be noted that the term "including" and its variations used in the embodiments of the present invention are open-ended, i.e., "including but not limited to." The term "based on" means "based at least in part on." The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; and the term "some embodiments" means "at least some embodiments." The modifications of "one" and "a plurality of" mentioned in the embodiments of the present invention are illustrative and non-restrictive. Those skilled in the art should understand that, unless the context clearly indicates otherwise, they should be understood as "one or more."

[0179] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0180] The various steps described in the method implementation scheme provided in the embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method implementation scheme may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0181] The term "embodiment" in this specification refers to specific features, structures, or characteristics described in conjunction with the embodiment that can be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. The various embodiments in this specification are described in a related manner, and the same or similar parts between the various embodiments are referenced to each other. In particular, for the embodiments of the device, equipment, and system, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts are referred to the partial description of the method embodiment.

[0182] The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of the present invention. Therefore, the scope of the present invention shall be determined by the appended claims.

Claims

1. A SAR image enhancement method based on regional adaptation, characterized in that: include: Performing discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image; Based on the initial multi-scale wavelet coefficients, an initial universal hidden Markov tree model corresponding to the initial ocean SAR image is established, and a high-noise region image and a low-noise region image of the initial ocean SAR image are determined, wherein the initial universal hidden Markov tree model is used to determine a hidden state probability set of the initial ocean SAR image; Performing discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image; Based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively establishing a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image, the high-noise universal hidden Markov tree model being used to determine a hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model being used to determine a hidden state probability set of the low-noise region image; Determining the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and determining the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image; updating the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients to obtain an updated target universal hidden Markov tree model; Performing inverse wavelet transform on the wavelet coefficients in the target universal hidden Markov tree model to obtain an enhanced target ocean SAR image.

2. The method according to claim 1, characterized in that The initial multi-scale wavelet coefficients include a first high-frequency characteristic wavelet coefficient and a first low-frequency characteristic wavelet coefficient, the high-noise multi-scale wavelet coefficients include a second high-frequency characteristic wavelet coefficient and a second low-frequency characteristic wavelet coefficient, and the low-noise multi-scale wavelet coefficients include a third high-frequency characteristic wavelet coefficient and a third low-frequency characteristic wavelet coefficient.

3. The method according to claim 2, characterized in that The performing discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image includes: Based on the Haar wavelet basis function, the initial ocean SAR image is processed by a high-pass filter and a low-pass filter respectively to generate the first high-frequency characteristic wavelet coefficient and the first low-frequency characteristic wavelet coefficient.

4. The method according to claim 2, characterized in that The step of establishing an initial general hidden Markov tree model corresponding to the initial ocean SAR image based on the initial multi-scale wavelet coefficients and determining a high-noise region image and a low-noise region image of the initial ocean SAR image comprises: Determining a first probability density function of a mixed Gaussian model of each of the initial multi-scale wavelet coefficients, and determining a first node relationship between different scale layers of each of the initial multi-scale wavelet coefficients through a parameter set of a hidden Markov tree model; Establishing the initial universal hidden Markov tree model based on the first probability density function and the first node relationship; The first probability density function is determined based on the following formula (1): （1）； in, is the initial multi-scale wavelet coefficient; for The first probability density function of ; is the number of Gaussian distribution components in the mixed Gaussian model; Status The prior probability of ; for Given The probability of a Gaussian distribution component, For the The mean vector of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution; The parameter set of the hidden Markov tree model is determined based on the following formula (2): （2）； in, is the parameter set of the hidden Markov tree model; Root node status The state probability distribution function indicates that the root node is in the hidden state The probability of the following; For the parent node and child nodes The state transition probability between the parent node and the parent node is When the child node transitions to state probability; For the current node In the implicit state The mean of the probability density function of the wavelet coefficients under ; For the current node In the implicit state The variance of the probability density function of the wavelet coefficients under ; The implicit state probability set of the initial ocean SAR image is determined based on the following method: The hidden state probability set is determined by using a maximum expectation algorithm based on the meta-parameters of the initial universal hidden Markov tree model.

5. The method according to claim 2, characterized in that The step of establishing an initial general hidden Markov tree model corresponding to the initial ocean SAR image based on the initial multi-scale wavelet coefficients and determining a high-noise region image and a low-noise region image of the initial ocean SAR image comprises: Performing sliding window processing on the initial ocean SAR image based on a window of a preset size, and determining the median value of the first scale HH subband of multiple initial multi-scale wavelet coefficients in each window; The median values ​​corresponding to each window are sorted, and a sorting result is output, so as to determine the high-noise region image and the low-noise region image based on the sorting result.

6. The method according to claim 2, characterized in that The method of establishing a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively, comprises: Determine a second probability density function of each of the high-noise multiscale wavelet coefficient and the low-noise multiscale wavelet coefficient; Determining, by means of a parameter set of a hidden Markov tree model, a second node relationship between the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients at different scale layers; Based on the second probability density function and the second node relationship, the high-noise universal hidden Markov tree model and the low-noise universal hidden Markov tree model are respectively established.

7. The method according to claim 2, characterized in that The step of determining the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and determining the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image, comprises: The conditional means corresponding to the high-noise region image and the low-noise region image are determined based on the following formula (3): （3）； in, Wavelet coefficients representing the signal in the image; Represents the wavelet coefficients of the parent node at different scales; Representation node The implicit state of A specific value representing an implicit state; Representation node In the implicit state The variance of the wavelet coefficients under ; represents the variance of the noise; Representation node The parent node wavelet coefficient of ; Based on the implicit state probability set of the high-noise region image and the implicit state probability set of the low-noise region image, the wavelet coefficients of the high-noise region and the wavelet coefficients of the low-noise region are respectively determined based on the following formula (4): （4）； in, represents the new wavelet coefficients, The wavelet coefficients representing the signal in a local image are high-noise region images or low-noise region images.

8. A marine synthetic aperture radar image enhancement device based on regional adaptation, characterized in that: include: A generating module, configured to perform discrete wavelet transform processing on the initial ocean SAR image to generate initial multi-scale wavelet coefficients corresponding to the initial ocean SAR image; a processing module, configured to establish an initial universal hidden Markov tree model corresponding to the initial ocean SAR image based on the initial multi-scale wavelet coefficients, and determine a high-noise region image and a low-noise region image of the initial ocean SAR image, wherein the initial universal hidden Markov tree model is used to determine a hidden state probability set of the initial ocean SAR image; The generating module is further configured to perform discrete wavelet transform processing on the high-noise region image and the low-noise region image respectively to generate high-noise multiscale wavelet coefficients corresponding to the high-noise region image and low-noise multiscale wavelet coefficients corresponding to the low-noise region image; An establishment module is used to establish a high-noise universal hidden Markov tree model corresponding to the high-noise region image and a low-noise universal hidden Markov tree model corresponding to the low-noise region image based on the high-noise multiscale wavelet coefficients and the low-noise multiscale wavelet coefficients, respectively, wherein the high-noise universal hidden Markov tree model is used to determine the hidden state probability set of the high-noise region image, and the low-noise universal hidden Markov tree model is used to determine the hidden state probability set of the low-noise region image; a determination module, configured to determine the wavelet coefficients of the high-noise region according to the implicit state probability set of the high-noise region image, and to determine the wavelet coefficients of the low-noise region according to the implicit state probability set of the low-noise region image; an updating module, configured to update the initial multi-scale wavelet coefficients corresponding to the high-noise region image and the low-noise region image in the initial universal hidden Markov tree model based on the high-noise region wavelet coefficients and the low-noise region wavelet coefficients, to obtain an updated target universal hidden Markov tree model; The transformation module is used to perform inverse wavelet transformation on the wavelet coefficients in the target universal hidden Markov tree model to obtain the target ocean SAR image after image enhancement.

9. An electronic device comprising: A processor and a memory storing a program, wherein the program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory machine-readable medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 7.

Images