1. A SAR image superpixel sliding window CFAR detection method is characterized by comprising the following steps: the method comprises the following steps:

step 1: setting a superpixel size S, and performing superpixel segmentation on the SAR image to be detected by adopting a simple linear iterative clustering SLIC method to obtain a superpixel { S_m1,2, …, M, wherein M is the number of superpixels;

step 2: setting a local sliding window consisting of a to-be-detected super pixel and a background super pixel, adaptively calculating a clutter truncation threshold through a clutter statistical probability histogram in the background super pixel, removing heterogeneous pixel points of a target and a fuzzy azimuth leaked into the background super pixel, and reserving a real clutter sample;

and step 3: modeling clutter after adaptive threshold truncation in background super pixels by adopting truncation gamma distribution, and obtaining clutter parameters according to a truncation moment estimation method;

and 4, step 4: according to clutter parameters obtained by the estimation method, a false alarm probability is given, a truncated gamma distribution CFAR detection threshold value is solved in a self-adaptive mode, target discrimination is conducted on pixels to be detected in the superpixels to be detected, target discrimination is completed on the pixels in all M superpixels to be detected, and therefore SAR image target detection is achieved.

2. The SAR image superpixel sliding window CFAR detection method according to claim 1, characterized in that: the method for calculating the adaptive clutter truncation threshold t according to the clutter statistical histogram in the background super pixel in the step 2 comprises the following steps: firstly, counting a probability histogram P of gray values of pixel points in background super pixels, wherein P is more than or equal to 0 and less than or equal to P (i) and less than or equal to 1,wherein H is the gray level number, P (i) represents the probability value of ith gray level, and the gray level H where the maximum value of the probability histogram is located is searched_maxCalculating clutter truncation threshold as t ═ H_max+ H)/2H, and setting the gray value of a certain pixel in the background super pixel as I_BThe truncation rule is I_BAnd (5) keeping the clutter of which the gray value is less than t as the true sea clutter when the gray value is less than t.

3. The SAR image superpixel sliding window CFAR detection method according to claim 1, characterized in that: modeling the clutter after adaptive threshold truncation in the background super-pixel by adopting truncation gamma distribution, and obtaining clutter parameters according to a truncation moment estimation method, wherein the method comprises the following specific steps: using the following gamma distribution to the clutter samples truncated by the adaptive threshold tThe gray level probability density of (a):

in the formula (I), the compound is shown in the specification,l and μ are the shape parameter and mean of the gamma distribution, respectivelyParameters by computing first order moments of truncated clutter samplesAnd second momentThe following simultaneous equations are established:

wherein the first and second moments of the truncated clutter samples are passedAndin an approximate representation of the above-described,n is the number of truncated clutter samples toIterative solution as an initial point to obtain a numerical solutionWherein

4. The SAR image superpixel sliding window CFAR detection method according to claim 1, characterized in that: the method comprises the following steps of self-adaptively solving a truncated gamma distribution CFAR detection threshold value in the step 4, carrying out target discrimination on pixel points in the superpixels to be detected, and finishing target discrimination on the pixel points in all M superpixels to be detected, wherein the specific method comprises the following steps:setting the mth super pixel S to be detected_mThe clutter parameter estimation result isGiven false alarm probability P_faThe CFAR detection threshold T is calculated according to the following formula_m：

By T_mSuper pixel S to be detected_mGray value of K pixel points { I }_m1,I_m2,…,I_mk,…,I_mKJudging, the CFAR detection target judgment rule is I_mk≥T_m。

5. The SAR image superpixel sliding window CFAR detection method according to claim 1, characterized in that: the background superpixel in the superpixel local sliding window is determined by adopting the following method:

and drawing a circle by taking the mass center of the super pixel to be detected as the circle center and the size s of the super pixel as the radius, and selecting all the super pixels covered in the circle as background super pixels.

Background

As an active microwave sensor, a Synthetic Aperture Radar (SAR) system is not limited by conditions such as illumination, weather and the like, has all-weather and all-day observation capability, and is a research hotspot in the field of radar detection at present by utilizing SAR images to detect sea and ground targets.

In the existing SAR image target detection method, the double-parameter constant false alarm CFAR detection method is widely applied. The detection algorithm is based on the assumption that background clutter obeys Gaussian or lognormal distribution, a sliding window consisting of a target window, a protection window and a background window is set, and pixels in the whole SAR image are traversed by utilizing the sliding window. In the traditional CFAR method, a protection window is used for preventing partial pixels of a target from leaking to a background window under a complex background and influencing the accuracy of clutter parameter estimation, however, in a busy shipping area such as a port, the interference of heterogeneous pixels such as adjacent target pixels and azimuth ambiguity cannot be completely eliminated only by adopting the protection window. Therefore, the drawbacks of the two-parameter CFAR method are mainly: background clutter modeling is inaccurate in complex scenes such as multiple targets, so that the target detection accuracy is reduced; target detection is realized by adopting a pixel point sliding window, and when detection judgment is carried out on each pixel point, background clutter parameters are required to be estimated, so that the algorithm speed is slow.

At present, aiming at the problem of reduction of detection rate in complex environments such as multiple targets, a CFAR detection method based on sample screening is provided, parameter estimation and statistical modeling are carried out on clutter screened out in an iteration mode, target detection precision is effectively improved, however, clutter screening is carried out by adopting a fixed threshold value, so that real clutter samples are discarded, clutter parameter estimation precision is reduced, and the calculation efficiency of a pixel point sliding window iteration screening mode is low.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a SAR image superpixel sliding window CFAR detection method based on truncated gamma clutter, which adopts superpixels as basic units of CFAR detection, constructs a superpixel sliding window containing superpixels to be detected and background superpixels, adopts a self-adaptive threshold value to truncate and screen a real clutter sample in the background superpixels, utilizes truncated gamma distribution to carry out accurate clutter modeling, and calculates a CFAR detection threshold value according to a set false alarm probability to obtain a target detection result.

A SAR image superpixel sliding window CFAR detection method based on truncated gamma clutter comprises the following steps:

step 1: setting a superpixel size S, and performing superpixel segmentation on the SAR image to be detected by adopting a simple linear iterative clustering SLIC method to obtain a superpixel { S_m1,2, …, M, wherein M is the number of superpixels;

step 2: setting a local sliding window consisting of a to-be-detected super pixel and a background super pixel, adaptively calculating a clutter truncation threshold through a clutter statistical probability histogram in the background super pixel, removing heterogeneous pixel points of a target and a fuzzy azimuth leaked into the background super pixel, and reserving a real clutter sample;

preferably, the background superpixel in the superpixel local sliding window is determined by the following method:

and drawing a circle by taking the mass center of the super pixel to be detected as the circle center and the size s of the super pixel as the radius, and selecting all the super pixels covered in the circle as background super pixels.

Preferably, the adaptive clutter truncation threshold calculation method is as follows:

firstly, counting a probability histogram P of gray values of pixel points in background super pixels, wherein P is more than or equal to 0 and less than or equal to P (i) and less than or equal to 1,wherein H is the gray level number, P (i) represents the probability value of ith gray level, and the gray level H where the maximum value of the probability histogram is located is searched_maxCalculating clutter truncation threshold as t ═ H_max+ H)/2H, and setting the gray value of a certain pixel in the background super pixel as I_BThe truncation rule is I_BAnd (5) keeping the clutter of which the gray value is less than t as the true sea clutter when the gray value is less than t.

And step 3: modeling the clutter after adaptive threshold truncation in the background super-pixel by adopting truncation gamma distribution, and obtaining clutter parameters according to a truncation moment estimation method, wherein the method specifically comprises the following steps:

using the following gamma distribution to the clutter samples truncated by the adaptive threshold tIs constructed by the gray scale probability densityDie:

in the formula (I), the compound is shown in the specification,l and mu are respectively the shape parameter and the mean value parameter of gamma distribution, and the first moment of the truncated clutter sample is calculatedAnd second momentThe following simultaneous equations are established:

wherein the first and second moments of the truncated clutter samples are passedAndin an approximate representation of the above-described,n is the number of truncated clutter samples toIterative solution as an initial point to obtain a numerical solutionWherein

And 4, step 4: according to clutter parameters obtained by the estimation method, a false alarm probability is given, a truncated gamma distribution CFAR detection threshold value is solved in a self-adaptive mode, target discrimination is conducted on pixels to be detected in the superpixels to be detected, target discrimination is completed on the pixels in all M superpixels to be detected, and therefore SAR image target detection is achieved.

The method comprises the following specific steps:

setting the mth super pixel S to be detected_mThe clutter parameter estimation result isGiven false alarm probability P_faThe CFAR detection threshold T is calculated according to the following formula_m：

By T_mSuper pixel S to be detected_mGray value of K pixel points { I }_m1,I_m2,…,I_mk,…,I_mKJudging, the CFAR detection target judgment rule is I_mk≥T_m。

Compared with the prior art, the invention has the remarkable advantages that:

1. the method can adaptively calculate the truncation threshold, and reduce clutter fitting deviation brought by the fixed threshold;

2. according to the method, the truncated clutter is statistically modeled by adopting gamma distribution, and compared with truncated log-normal distribution, the clutter fitting precision is improved;

3. the method realizes target detection by using the superpixel sliding window, and greatly improves the computational efficiency of CFAR detection.

Drawings

FIG. 1 is a diagram of a superpixel sliding window architecture;

FIG. 2 is a comparison result of fitting the truncated clutter gray level histogram to the truncated filtered clutter gray level histogram in the truncated gamma model and the truncated log-normal model in the comparison method according to the present invention;

FIG. 3 is an original SAR image to be detected, wherein a square block is marked with a target true value;

FIG. 4 is a graph comparing the results of target detection by the method of the present invention with those of the prior art;

FIG. 5 is a graph comparing ROC curves for the method of the present invention and a prior art method.

Detailed Description

The following will further explain the steps and effects of the present invention with reference to the drawings.

Step 1: setting a superpixel size S, and performing superpixel segmentation on the SAR image to be detected by adopting a simple linear iterative clustering SLIC method to obtain a superpixel { S_m1,2, …, M, wherein M is the number of superpixels;

step 2: as shown in the attached figure 1, a local sliding window consisting of a to-be-detected super pixel and a background super pixel is set, a clutter truncation threshold is adaptively calculated through a clutter statistical histogram in the background super pixel, heterogeneous pixel points such as targets and azimuth ambiguities leaked into the background super pixel are removed, and a real clutter sample is reserved.

Specifically, the background superpixel in the superpixel local sliding window is determined by the following method:

with a super-pixel S to be detected_mCenter of mass C_mM is 1,2, …, M is the center of a circle, the superpixel size s is used as the radius to draw a circle, all superpixels covered in the circle are selected as background superpixels, and specifically, the selection rule is as follows: the selected background superpixel set is S_kIs satisfied withAnd k ∈ [1, M ]]K is not equal to m, wherein C_kIs a super pixel S_kThe center of mass of the lens.

Specifically, the adaptive clutter truncation threshold calculation method is as follows:

firstly, counting a probability histogram P of gray values of pixel points in background super pixels, wherein P is more than or equal to 0 and less than or equal to P (i) and less than or equal to 1,wherein H is the gray level number, searching the gray level H where the maximum value of the probability histogram P is_maxCalculating clutter truncation threshold as t ═(H_max+ H)/2H, and setting the gray value of a certain pixel in the background super pixel as I_BThe truncation rule is I_BAnd (5) keeping the clutter of which the gray value is less than t as the true sea clutter when the gray value is less than t.

And step 3: modeling the clutter after adaptive threshold truncation in the background super-pixel by adopting truncation gamma distribution, and obtaining clutter parameters according to a truncation moment estimation method.

Specifically, the following gamma distribution is adopted for clutter samples cut by adopting an adaptive threshold value tThe gray level probability density of (a):

where L and μ are the shape parameter and mean parameter of the gamma distribution, respectively.

In particular, the clutter parameters L and μmay be calculated by computing the first moment of truncated clutter samplesAnd second momentSolving the following equation set to obtain a numerical solution

Wherein the first and second moments of the truncated clutter samples are passedAndin an approximate representation of the above-described,n is the number of truncated clutter samples toIterative solution as an initial point to obtain a numerical solutionWherein

And 4, step 4: according to the given false alarm probability, a cut gamma distribution CFAR detection threshold is solved in a self-adaptive mode, the target discrimination is carried out on the pixel points in the superpixels to be detected, the target discrimination is completed on the pixel points in all M superpixels to be detected, and therefore SAR image target detection is achieved.

Setting the mth super pixel S to be detected_mThe clutter parameter estimation result isGiven false alarm probability P_faThe CFAR detection threshold T is calculated according to the following formula_m：

By T_mSuper pixel S to be detected_mGray value of K pixel points { I }_m1,I_m2,…,I_mk,…,I_mKJudging, the CFAR detection target judgment rule is I_mk≥T_m。

So far, the superpixel sliding window CFAR detection method based on truncated gamma clutter estimation is basically completed.

The effectiveness of the invention is further illustrated by a Sentinel-1 satellite SAR image target detection contrast experiment.

1. Experimental setup:

the experimental data were from SAR data from a Sentinel-1 satellite imaging the panama region 12 months 2015 with range and azimuth resolutions of about 12m and 14m, respectively, and an image size of 730 x 700 pixels. True target values are marked in the figure with white boxes. In the experiment, the detection performance of the CFAR (AC-CFAR), the superpixel CFAR (SP-CFAR) and the truncated log normal distribution CFAR (TS-LNCFAR) is compared with the truncated gamma superpixel CFAR (TSSP-CFAR) provided by the invention.

The sizes of the target window, the protection window, and the background window for AC-CFAR are set to 1 × 1, 41 × 41, and 81 × 81, respectively, and the pixel target confidence in the initial detection is set to 3%. For TS-LNCFAR, the target and background windows are set to 1 × 1 and 81 × 81, respectively, the iterative filtering number is set to 5, and the truncation depth factor is set to 2. For the SP-CFAR and TSSP-CFAR methods of the present invention, the superpixel size is set at s-25, and the superpixel clutter window structure employed by the methods of the present invention is shown in fig. 1.

2. And (4) analyzing results:

in the experiment, KL distance and histogram fitting results are adopted to explain the effectiveness of the truncated gamma distribution in fitting clutter parameters, and an ROC curve and detection efficiency are adopted to carry out quantitative analysis on the method and the comparison method, wherein the horizontal axis of the ROC curve is the false alarm rate FAR of the detection result, and the vertical axis of the ROC curve is the detection rate DR of the detection result. Firstly, the fitting result of truncated gamma and truncated log-normal distribution to the gray level histogram of truncated screening clutter is shown in fig. 2, so that the fitting effect of truncated gamma distribution is better, quantitative verification can be obtained from the KL distance in table 1, and the KL distance between the truncated gamma distribution and the gray level histogram of truncated clutter is smaller, which indicates that the fitting effect is better; as shown in fig. 3, the original SAR image to be detected is shown, wherein a square block is marked as a target true value;

TABLE 1 comparison of fitting results of different models to truncated clutter

Model (model)	Truncated log normal	Truncation Gamma
			KL distance	0.088	0.019

TABLE 2 comparison of target detection efficiency of the inventive and comparative methods

Method	AC-CFAR	TS-LNCFAR	SP-CFAR	TS-SPCFAR
					Detection time(s)	11.3	175	1.1	2.6

When the actual false alarm rate is 0.02%, the pair of target detection results of AC-CFAR, SP-CFAR, TS-LNCFAR and TSSP-CFAR proposed by the present invention is shown in fig. 4, where the dotted line box indicates that the target structure in the detection result is missing, and the solid line box indicates that there is missing detection, and it can be seen from the visual comparison that there is no missing detection in the detection result of the TSSP-CFAR method proposed by the present invention, and the detection accuracy is the highest, which can be quantitatively verified from the ROC performance curve shown in fig. 5, and under the same false alarm rate, the detection rate of the method of the present invention is significantly higher than that of other comparison methods. In addition, as can be seen from the comparison result of the target detection efficiency in table 2, the detection time of the method of the present invention is second to that of SP-CFAR, and the requirements of practical engineering applications can be met.

In conclusion, the SAR image superpixel sliding window CFAR detection method based on the truncated gamma clutter can improve clutter modeling precision under complex backgrounds such as multiple targets, keeps a high detection rate while maintaining a low false alarm rate, and is favorable for practical engineering application due to good detection efficiency.