A fast maximum likelihood estimation method for equivalent number of looks map of polarimetric SAR image
By combining logarithmic determinant graphs and convolution operations, the equivalent number of views of polarimetric SAR images can be estimated quickly, solving the problem of excessive time consumption in existing technologies and achieving efficient number of views calculation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-23
- Publication Date
- 2026-03-27
AI Technical Summary
Existing maximum likelihood estimation methods for equivalent number of views of polarimetric SAR images lack analytical expressions, resulting in excessive computation time and making it difficult to meet the needs of high-efficiency applications.
A method combining logarithmic determinant plots and convolution operations is adopted. By calculating the logarithmic determinant plot, local mean plot, and local sample statistics plot of the polarimetric SAR image, the equivalent number of views plot is quickly estimated using the analytical approximation of ML estimation.
It significantly improves the computational efficiency of equivalent number of views maps for polarimetric SAR images, enabling fast and accurate estimation and meeting the needs of high-efficiency applications.
Smart Images
Figure CN116736300B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of imaging radar image processing, more particularly to a fast implementation method of Maximum Likelihood (ML) estimation of a polarimetric Synthetic Aperture Radar (SAR) image equivalent number of looks map. BACKGROUND
[0002] Polarimetric SAR is a kind of SAR with multiple polarization transmitting and receiving combination modes, which has strong information acquisition capability, and can work all day and all weather due to the active microwave imaging mode, which is not affected by weather factors such as clouds, rain and fog. This makes the polarimetric SAR image processing technology have important application value in both civil and military fields.
[0003] Due to the coherent imaging mechanism of SAR, a large amount of random speckle noise is contained in the polarimetric SAR image, which greatly increases the difficulty of interpretation, and also makes the method based on statistical distribution model an important way for polarimetric SAR image interpretation. The complex Wishart distribution is the most widely used statistical distribution model for polarimetric SAR image, and the equivalent number of looks is a key parameter, whose accuracy directly affects the reliability of the distribution model. Physically, the equivalent number of looks reflects the degree of averaging of polarimetric SAR image data and can measure the uniformity of the corresponding data. The polarimetric SAR image equivalent number of looks map is an image composed of the equivalent number of looks corresponding to the neighborhood data of each pixel, which reflects the spatial data uniformity distribution of the polarimetric SAR image, and is widely used in polarimetric SAR image filtering, segmentation and classification algorithms. Therefore, it is of great significance to estimate the polarimetric SAR image equivalent number of looks map quickly and effectively to promote the interpretation of polarimetric SAR image.
[0004] Over the past few decades, numerous methods for estimating the equivalent number of views (OLDs) of polarimetric SAR images have been proposed, including methods of moment estimation, fractional-order variance coefficient method, trace moment estimation method, and ML estimation method. Variance coefficient method and fractional-order moment estimation method primarily rely on single-channel polarimetric SAR image data for estimation. While easy to implement, these methods utilize only the diagonal elements of the polarization matrix (polarization covariance matrix or polarization coherence matrix), resulting in insufficient effectiveness. In contrast, trace moment estimation method and ML estimation method utilize all elements of the polarization matrix, achieving more accurate OLDs. Compared to trace moment estimation method, ML estimation method exhibits lower bias and greater effectiveness, making it a crucial method for current OLDs estimation. However, the ML estimation method lacks an analytical expression and requires numerical solutions. This presents several practical problems: Firstly, the method is time-consuming due to the need for numerous iterative numerical calculations to approximate the solution. Secondly, a suitable range for the equivalent number of views needs to be determined before the solution is obtained. If this range is too narrow and does not include the true solution, reliable estimates will be difficult to obtain. If the range is too large, this will further increase the computation time. In particular, when estimating the equivalent number of views of a polarimetric SAR image, it is necessary to estimate the equivalent number of views corresponding to each pixel. Since ML estimation lacks an analytical expression, it is not conducive to parallel implementation and uses a pixel-by-pixel estimation method. Therefore, its implementation is very time-consuming and cannot meet the needs of many practical applications with high timeliness requirements. Summary of the Invention
[0005] The technical problem to be solved by this invention is:
[0006] To address the problem that existing ML estimation of equivalent look number maps of polarimetric SAR images lacks analytical solutions and is extremely time-consuming in practical applications, a fast estimation method is proposed that can be implemented while ensuring estimation accuracy, so as to better meet the needs of a large number of practical applications with high timeliness requirements.
[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0008] A fast maximum likelihood estimation method for equivalent number of views map of polarimetric SAR image, characterized by the following steps:
[0009] Step 1: Calculate the logarithmic determinant of the polarimetric SAR image
[0010] The determinant values of the polarization matrix of all pixels in the polarimetric SAR image are calculated based on matrix operations to construct the corresponding determinant graph; the logarithmic determinant graph of the polarimetric SAR image data is obtained by logarithmic operations.
[0011] Step 2: Calculate the local mean map of the logarithmic determinant of the polarimetric SAR image.
[0012] Given the pixel neighborhood window size, the local mean map of the log-determinant of the polarimetric SAR image is calculated by using the convolution method.
[0013] Step 3: Calculate the local mean map of each channel data of the polarimetric SAR image
[0014] Given the pixel neighborhood window size, the local mean map of each channel data of the polarimetric SAR image is calculated by using the convolution method.
[0015] Step 4: Calculate the log-determinant map of the local average polarimetric SAR image
[0016] Based on the matrix point multiplication and matrix summation operation, the determinant value of the polarization matrix of all pixels of the local average polarimetric SAR image is calculated, and the corresponding determinant map is constructed; the corresponding log-determinant map is obtained by using the logarithmic operation;
[0017] Step 5: Calculate the local sample statistical quantity map of the polarimetric SAR image
[0018] According to the local mean map of the log-determinant of the polarimetric SAR image in step 2 and the log-determinant map of the local average polarimetric SAR image in step 4, the local sample statistical quantity map required for the estimation of the equivalent number of views of the polarimetric SAR image is calculated.
[0019] Step 6: Estimate the approximate equivalent number of view map of the polarimetric SAR image
[0020] The equivalent number of view map of the polarimetric SAR image is quickly estimated by using the analytical approximation solution of the ML estimation.
[0021] Further technical solutions of the application: step 1 uses logarithmic operation to obtain the log-determinant map of the polarimetric SAR image data, which is specifically
[0022]
[0023]
[0024]
[0025] wherein, wherein Z m , m = 1, 2,..., 9 represent the mth element of the polarization matrix z, i.e. each pixel of the polarimetric SAR image, and represents the point multiplication operation of the matrix, i.e. the multiplication of the elements at the corresponding positions of two matrices of the same size.
[0026] Further technical solutions of the application: step 2 uses the convolution method to calculate the local mean map of the log-determinant of the polarimetric SAR image, which is specifically:
[0027]
[0028]
[0029]
[0030] where S1xS2 is the size of the pixel neighborhood window, Conv(·,·) represents the convolution operation of the matrix, and k is a convolution kernel with a size of S1xS2 and elements of 1 / (S1xS2).
[0031] A further technical solution of the present application is that step 3 adopts a convolution method to calculate the local mean value map of each channel data of the polarimetric SAR image, and is specifically as follows:
[0032]
[0033]
[0034] A further technical solution of the present application is that step 4 adopts logarithmic operation to obtain a corresponding logarithmic determinant map, and is specifically as follows:
[0035]
[0036]
[0037] A further technical solution of the present application is that step 5 calculates the local sample statistic map of the polarimetric SAR data, and is specifically as follows:
[0038]
[0039] A further technical solution of the present application is that step 6 estimates the approximate equivalent number of views map of the polarimetric SAR image, and is specifically as follows: A0=4G, A1=6(3-2G), A2=8G-37,
[0040]
[0041]
[0042]
[0043] where the formula The point division of the matrix A and the matrix B is expressed;
[0044] Then, the equivalent number of views map of the polarimetric SAR image can be obtained by solving the formula according to the following formula, which is an analytical approximation of the ML estimation of the equivalent number of views:
[0045]
[0046] An error estimation method of a fast maximum likelihood estimation method of a polarimetric SAR image equivalent number of views map is characterized in that a relative estimation error REE is defined:
[0047]
[0048] wherein, and REE is the relative error of the equivalent number of looks of the polarimetric SAR image obtained by the method of claim 1 and the conventional ML estimation method, expressed in percentage; the smaller the value of REE, the closer the result obtained by the method of claim 1 to the result obtained by the conventional ML method.
[0049] A computer system, characterized in comprising: one or more processors, a computer readable storage medium storing one or more programs for execution by the one or more processors, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the above method.
[0050] A computer readable storage medium storing computer executable instructions, which when executed cause a computer to implement the above method.
[0051] The present application has the following beneficial effects:
[0052] The present application provides a fast maximum likelihood estimation method for the equivalent number of looks of a polarimetric SAR image, which uses the asymptotic form of the Digamma function to obtain an approximate analytical solution for the ML estimation of the equivalent number of looks of a polarimetric SAR image, and uses matrix operations and convolution operations to realize fast calculation of the local sample statistics of the equivalent number of looks of polarimetric SAR data, thereby effectively improving the implementation efficiency of the ML estimation of the equivalent number of looks of a polarimetric SAR image. The method of the present application has important supporting value for high timeliness interpretation of polarimetric SAR images. BRIEF DESCRIPTION OF DRAWINGS
[0053] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:
[0054] Figure 1 A flowchart of the fast ML estimation method for the equivalent number of looks of a polarimetric SAR image of the present application;
[0055] Figure 2 A Pauli-RGB pseudo-color image of a polarimetric SAR image used in the present application, which is constructed by taking the diagonal elements of the polarimetric coherence matrix as the red, green and blue basic color components, respectively.
[0056] Figure 3Logarithmic display results of equivalent view count maps of polarimetric SAR images estimated by the conventional ML method and the method of the present invention under different neighborhood window sizes: (a) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the conventional ML estimation method under a 3×3 neighborhood window condition; (b) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the method of the present invention under a 3×3 neighborhood window condition; (c) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the conventional ML estimation method under a 7×7 neighborhood window condition; (d) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the method of the present invention under a 7×7 neighborhood window condition; (e) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the conventional ML estimation method under an 11×11 neighborhood window condition; (f) Logarithmic display result of equivalent view count map of polarimetric SAR image obtained by the method of the present invention under an 11×11 neighborhood window condition.
[0057] Figure 4 for Figure 3 Histograms of equivalent number of views for each polarimetric SAR image; (a) Histograms of equivalent number of views for polarimetric SAR images obtained by the traditional ML estimation method and the method of the present invention under a 3×3 neighborhood window condition; (b) Histograms of equivalent number of views for polarimetric SAR images obtained by the traditional ML estimation method and the method of the present invention under a 7×7 neighborhood window condition; (c) Histograms of equivalent number of views for polarimetric SAR images obtained by the traditional ML estimation method and the method of the present invention under an 11×11 neighborhood window condition;
[0058] Figure 5 Histograms of the relative error between the equivalent number of views of polarimetric SAR images obtained by the traditional ML method and the method of this invention are shown below: (a) Histogram of the relative error between the equivalent number of views of polarimetric SAR images obtained by the traditional ML estimation method and the method of this invention under a 3×3 neighborhood window condition; (b) Histogram of the relative error between the equivalent number of views of polarimetric SAR images obtained by the traditional ML estimation method and the method of this invention under a 7×7 neighborhood window condition; (c) Histogram of the relative error between the equivalent number of views of polarimetric SAR images obtained by the traditional ML estimation method and the method of this invention under an 11×11 neighborhood window condition. Detailed Implementation
[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0060] The embodiment of the application provides a fast maximum likelihood estimation method for a polarized SAR image equivalent view number graph, wherein a Digamma function contained in an original ML estimation equation of a polarized SAR image equivalent view number is replaced by an asymptotic form of the Digamma function, and a high-order term in the equation is discarded to derive an ML estimation approximate equation with an analytical expression, so that iterative numerical operation and initial interval setting problems of a traditional method are avoided. In addition, a fast calculation method based on matrix operation and convolution is designed in terms of sample statistical quantity calculation, and the efficiency advantage is particularly obvious when the equivalent view number graph of the polarized SAR image is estimated.
[0061] Referring to the drawings Figure 1 In the scheme, firstly, based on matrix point multiplication and matrix summation operation, the determinant values of all pixel data of the polarized SAR image are calculated to form a corresponding determinant graph, and then a logarithmic operation is adopted to obtain a logarithmic determinant graph of the polarized SAR image data. Next, a corresponding convolution kernel is constructed by giving a pixel neighborhood window size, and a convolution method is adopted to calculate a local mean value graph of the polarized SAR logarithmic determinant graph. Next, a convolution method is similarly adopted to calculate a local mean value graph of each channel data of the polarized SAR. Subsequently, based on matrix point multiplication and matrix summation operation, the determinant values of all pixel data of the local average polarized SAR image are calculated to form a corresponding determinant graph, and then a logarithmic operation is adopted to obtain a logarithmic determinant graph thereof. Next, a matrix subtraction is adopted to obtain a local sample statistical quantity graph of the polarized SAR data equivalent view number estimation. Finally, the analytical approximate solution formula of the ML estimation is solved based on matrix operation, and the equivalent view number graph of the polarized SAR image can be quickly estimated. The method comprises the following steps:
[0062] Step one: calculating a logarithmic determinant graph of the polarized SAR image
[0063] Firstly, based on matrix operation, the determinant values of the polarization matrix of all pixels of the polarized SAR image are calculated to form a corresponding determinant graph. Then, a logarithmic operation is adopted to obtain a logarithmic determinant graph of the polarized SAR image data.
[0064] Step two: calculating a local mean value graph of the logarithmic determinant of the polarized SAR image
[0065] Given a pixel neighborhood window size, a convolution method is adopted to calculate a local mean value graph of the logarithmic determinant of the polarized SAR image.
[0066] Step three: calculating a local mean value graph of each channel data of the polarized SAR image
[0067] Given a pixel neighborhood window size, a convolution method is adopted to calculate a local mean value graph of each channel data of the polarized SAR image.
[0068] Step four: calculating a logarithmic determinant graph of the local average polarized SAR image
[0069] First, based on the matrix point multiplication and matrix summation operation, the determinant value of the polarization matrix of all pixels of the local average polarimetric SAR image is calculated, and the corresponding determinant map is constructed. Then the corresponding logarithmic determinant map is obtained by using logarithmic operation.
[0070] Step five: calculating the local sample statistic map of the polarimetric SAR image
[0071] According to the results of step two and step four, the local sample statistic map required for the estimation of the effective number of looks of the polarimetric SAR image is calculated.
[0072] Step six: estimating the approximate equivalent number of looks map of the polarimetric SAR image
[0073] The equivalent number of looks map of the polarimetric SAR image is quickly estimated by using the analytical approximation solution of the ML estimation.
[0074] The specific implementation steps of the present application are described in detail.
[0075] Step one: calculating the logarithmic determinant map of the polarimetric SAR image
[0076] Each pixel of the multi-look polarimetric SAR image is usually represented by a 3x3 polarization matrix (polarization covariance matrix or polarization coherence matrix) z, denoted as:
[0077]
[0078] wherein Z m , m = 1, 2,..., 9 represents the mth element of the polarization matrix z.
[0079] Given a polarimetric SAR image with a size of MxNx9, the data is denoted as wherein Z m , m = 1, 2,..., 9 is the data of the mth channel of the polarimetric SAR image, i.e. the MxN size matrix data composed of the mth element of the polarization matrix of each pixel of the polarimetric SAR image, denotes the splicing of the matrix.
[0080] The value of the determinant of all pixel data of the polarimetric SAR image is calculated according to the following formula, i.e. the corresponding determinant map is constructed:
[0081]
[0082] wherein represents the point multiplication operation of the matrix, i.e. the multiplication of the elements at the corresponding positions of two matrices with the same size.
[0083] Then, the logarithmic determinant map of the polarimetric SAR image data is obtained by using logarithmic operation
[0084] Step two: Calculate the local mean map of the log-determinant of the polarimetric SAR image
[0085] Given a pixel neighborhood window size S1xS2, construct a convolution kernel k with size S1xS2, whose elements are all 1 / (S1xS2), that is,
[0086]
[0087] Next, the polarimetric SAR log-determinant map Y is padded outwards, and (S1-1) / 2 and (S2-1) / 2 pixel data are filled in the two ends of the row direction and the column direction to obtain the expanded polarimetric SAR log-determinant map with size (M+S1-1)x(N+S2-1) denoted as:
[0088]
[0089] Convolution operation is performed on the expanded polarimetric SAR log-determinant map using the convolution kernel k to obtain the local mean map with size MxN:
[0090]
[0091] where Conv(·,·) represents the convolution operation of the matrix.
[0092] Step three: Calculate the local mean map of each channel data of the polarimetric SAR image
[0093] The each channel data z m ,m=1,2,...,9 of the polarimetric SAR image is padded outwards, and (S1-1) / 2 and (S2-1) / 2 pixel data are filled in the two ends of the row direction and the column direction to obtain the expanded polarimetric SAR image data with size (M+S1-1)x(N+S2-1):
[0094]
[0095] Convolution operation is performed on the expanded polarimetric SAR image data of each channel using the convolution kernel k to obtain the corresponding local mean map with size MxN:
[0096]
[0097] Further, the local average polarimetric SAR image data
[0098] Step four: Calculate the log-determinant map of the local average polarimetric SAR image
[0099] Based on matrix point multiplication and matrix summation operation, the determinant value of all pixel data of the local average polarimetric SAR image is calculated to form a corresponding determinant map:
[0100]
[0101] Then, the corresponding logarithmic determinant map is obtained by using logarithmic operation
[0102] Step five: calculating the local sample statistic map of the polarimetric SAR image
[0103] The local sample statistic map of the polarimetric SAR data is calculated by using the results of step two and step four:
[0104] Step six: estimating the approximate equivalent number of views map of the polarimetric SAR image
[0105] The following operations are performed on the local sample statistic G: A0=4G, A1=6(3-2G), A2=8G-37,
[0106]
[0107]
[0108]
[0109] wherein the formula The point division of matrix A and matrix B is expressed.
[0110] According to the following formula, that is, the analytical approximation of the ML estimation of the equivalent number of views, the equivalent number of views map of the polarimetric SAR image can be obtained:
[0111]
[0112] The effect of the present application can be further illustrated by the following simulation experiment:
[0113] 1. Simulation experiment content and result analysis
[0114] The present application will Figure 2 An L-band polarimetric SAR image of San Francisco area acquired by an AIRSAR system is taken as test data, as shown in the figure, and the size is 900*1024 pixels. In the test, three different sizes of pixel neighborhood windows are set, that is, 3*3, 7*7 and 9*9, and then the equivalent number of views map of the polarimetric SAR image under different neighborhood window conditions is obtained by using the traditional ML estimation method of the equivalent number of views and the method of the present application, wherein the results obtained by the traditional ML estimation method under the conditions of 3*3, 7*7 and 9*9 neighborhood windows are respectively as shown in the figures Figure 3As shown in (a), (c), and (e), the results obtained by the method of the present invention under the conditions of 3×3, 7×7, and 9×9 neighborhood windows are as follows: Figure 3 As shown in (b), (d), and (f). Additionally, histograms comparing these equivalent view counts under 3×3, 7×7, and 9×9 neighborhood window conditions are presented, as shown in... Figure 4 As shown in (a)-(c).
[0115] Depend on Figure 3 As can be seen, under the same conditions, the equivalent view number maps obtained by the traditional ML estimation method and the method of this invention look very similar, and their histograms are also very close, almost overlapping, indicating that the equivalent view number values estimated by the two methods are very close.
[0116] To quantitatively analyze the difference between the results obtained by these two methods, the following relative estimation error (REE) was defined in the test:
[0117]
[0118] in and These are the equivalent look counts of the polarimetric SAR images obtained by the method of this invention and the conventional ML estimation method, respectively. REE is expressed as a percentage, and its value reflects the relative error between the equivalent look counts obtained by the method of this invention and the equivalent look counts obtained by the conventional ML estimation method. The smaller the REE value, the closer the results obtained by the method of this invention and the conventional ML method are.
[0119] Under 3×3, 7×7, and 9×9 neighborhood window conditions, the histograms of REEs between the equivalent view maps obtained by the two methods are as follows: Figure 5 As shown in (a)-(c), the REE values are mainly distributed between 0.1% and 0.29, with all REE values less than 0.29%, making them extremely small and almost negligible. These results further verify that the method of this invention can obtain equivalent apparent values for SAR images that are extremely close to those obtained by traditional ML methods.
[0120] On the other hand, to evaluate the efficiency of the method of the present invention, the time consumed by the two methods in estimating the equivalent number of views of polarimetric SAR images was compared under 3×3, 7×7, and 9×9 neighborhood window conditions. For fair comparison, all experiments were run on a laptop with 32GB RAM and a 3.70GHz Intel Core i7-8700K CPU, and all were performed using MATLAB code. The experiments were repeated 10 times, and the time consumed by the two methods in calculating the sample statistics map (steps one to five) and solving the ML equation (step six) was obtained, and the total time was calculated, as shown in Table 1.
[0121] Table 1
[0122]
[0123] As shown in Table 1, the time required to calculate sample statistics increases with the increase in the neighborhood window size, which is due to the increased number of samples to be calculated. Traditional methods, employing pixel-by-pixel calculation, are very time-consuming: the calculation time for sample statistics maps is 27, 99, and 245 seconds for neighborhood window sizes of 3×3, 7×7, and 11×11, respectively. In contrast, the method of this invention uses parallel computation, achieving a significant efficiency improvement. Even with the longest time (less than 0.4 seconds) under an 11×11 neighborhood window, the time taken is still less than 0.4 seconds. Furthermore, the time taken by both the traditional method and the method of this invention to solve the ML equations is almost equal under different neighborhood window sizes, approximately 19 seconds and 0.3 seconds, respectively. Clearly, the method of this invention is more efficient than the traditional method, taking only about 1.6% of the time. In terms of total time, the traditional method is particularly time-consuming, especially when the neighborhood window is large, for example, approximately 260 seconds with an 11×11 window. In comparison, under the same conditions, the method of the present invention takes less than 0.8 seconds, which is less than 1 / 300th of the time taken by the traditional method.
[0124] Figure 3 The logarithmic results of the equivalent number of views of a polarimetric SAR image estimated by the conventional ML method and the method of this invention are displayed under different neighborhood window sizes (3×3, 7×7, and 11×11). The comparison shows that, under the same conditions, the equivalent number of views of the polarimetric SAR image data estimated by the two methods look very similar.
[0125] Figure 4 for Figure 3 Histograms of the equivalent number of views of polarimetric SAR image data estimated by the two methods are shown. It can be seen that, under the same conditions, the histograms of the equivalent number of views obtained by the two methods are very close, with only very subtle differences.
[0126] Figure 5 Histograms are shown for the relative error between the equivalent number of views of polarimetric SAR images obtained by the traditional ML method and the method of this invention under different neighborhood window sizes (3×3, 7×7, and 11×11). It can be seen that the relative error between the equivalent number of views estimated by the method of this invention and the result estimated by the standard traditional method is very small, with a maximum relative error of less than 0.3% (0.3%), which is almost negligible in practical applications. This demonstrates that the method of this invention can obtain equivalent number of views of polarimetric SAR images that are almost identical to those obtained by the standard method.
[0127] In conclusion, compared with the traditional ML method of the equivalent view number map of the polarimetric SAR image, the method of the application significantly improves the implementation efficiency on the basis of ensuring the estimation accuracy, and has important significance in high-efficiency practical applications.
[0128] The above merely describes a specific implementation of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the application, and these modifications or replacements shall be encompassed within the protection scope of the application.
Claims
1. A fast maximum likelihood estimation method of polarimetric SAR image equivalent number of looks map, characterized in that The steps are as follows: Step 1: calculating a log-determinant map of the polarimetric SAR image, The determinant values of the polarization matrix of all pixels of the polarimetric SAR image are calculated based on matrix operations to form a corresponding determinant map; and a log operation is used to obtain a log-determinant map of the polarimetric SAR image data; Step 2: calculating a local mean map of the log-determinant of the polarimetric SAR image, A local mean map of the log-determinant of the polarimetric SAR image is calculated using a convolution method given a pixel neighborhood window size; Step 3: calculating a local mean map of each channel data of the polarimetric SAR image, A local mean map of each channel data of the polarimetric SAR image is calculated using a convolution method given a pixel neighborhood window size; Step 4: calculating a log-determinant map of the polarimetric SAR image after local averaging, The determinant values of the polarization matrix of all pixels of the polarimetric SAR image after local averaging are calculated based on matrix point multiplication and matrix summation operations to form a corresponding determinant map; and a log operation is used to obtain a corresponding log-determinant map; Step 5: calculating a local sample statistic map of the polarimetric SAR image, According to the local mean map of the log-determinant of the polarimetric SAR image of step 2 and the log-determinant map of the polarimetric SAR image after local averaging of step 4, a local sample statistic map required for estimating the equivalent number of looks of the polarimetric SAR image is calculated; Step 6: estimating an approximate equivalent number of looks map of the polarimetric SAR image, the equivalent number of looks map of the polarimetric SAR image is quickly estimated using an analytical approximation of ML estimation; specifically: The following operations are performed on the local sample statistic G: , , , ， ， ， wherein the division of the matrix A by the matrix B is expressed by the matrix The expression point division of the matrix A by the matrix B; According to the following formula, i.e., the analytical approximation of ML estimation of the equivalent number of looks, the equivalent number of looks map of the polarimetric SAR image can be obtained: 。 2. The fast maximum likelihood estimation method of the equivalent number of looks map of polarimetric SAR images according to claim 1, characterized in that: Step 1: calculating a log-determinant map of the polarimetric SAR image data using a log operation is specifically ， ， ， wherein, is the data of the i-th channel of the polarimetric SAR image, m is the data of the i-th channel of the polarimetric SAR image, denotes the polarimetric matrix i.e. the i-th element of the pixel of the polarimetric SAR image, m i.e. the i-th element of the pixel of the polarimetric SAR image, denotes the dot product operation of matrices, i.e. the multiplication of the elements in the corresponding positions of two matrices of the same size.
3. The fast maximum likelihood estimation method of the equivalent number of looks map of polarimetric SAR images according to claim 2, characterized in that: Step 2: calculating a local mean map of the log-determinant of the polarimetric SAR image using a convolution method is specifically ， ， ， wherein, is a pixel neighborhood window size, denotes a convolution operation on a matrix, is a convolution kernel of size with elements .
4. The fast maximum likelihood estimation method of the equivalent number of looks map of polarimetric SAR images according to claim 3, characterized in that: Step 3: calculating a local mean map of each channel data of the polarimetric SAR image using a convolution method is specifically ， 。 5. The fast maximum likelihood estimation method of the equivalent number of looks map of polarimetric SAR images according to claim 4, characterized in that: Step 4: obtaining a corresponding log-determinant map using a log operation is specifically ， 。 6. The fast maximum likelihood estimation method of the equivalent number of looks map of polarimetric SAR images according to claim 5, characterized in that: Step 5 computes the local sample statistic map of polarimetric SAR data: .
7. A computer system, characterized by comprise: one or more processors, a computer readable storage medium storing one or more programs for execution by the one or more processors, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to carry out the method of claim 1.
8. A computer-readable storage medium, characterized in that computer executable instructions stored therein for implementing the method of claim 1 when executed.