CN113159157B - Improved low-frequency UWB SAR foliage covert target fusion change detection method - Google Patents

Abstract

The invention provides an improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method, which comprises the steps of firstly preprocessing a detection image and a reference image, and carrying out bidirectional relative radiation correction treatment; secondly, solving a detection quantity image and a detection threshold value based on an image segmentation difference method of the improved clutter distribution model; subtracting the detection image from the reference image to obtain a difference value detection quantity image, estimating clutter distribution of the difference value detection quantity image, setting false alarm probability to obtain a detection threshold value, and obtaining a detection quantity image and a detection threshold value of the first sub-algorithm; then, solving the inspection quantity images and the detection threshold of the one-dimensional edge worth method and the generalized Laguerre polynomial method; and finally, inputting the three inspection quantity images and the corresponding threshold values into an LE-SVDD space for designing a classifier, thereby realizing the change detection of the low-frequency UWB SAR image.

Description

Improved low-frequency UWB SAR foliage covert target fusion change detection method

technical field

The invention relates to the field of radar technology, and more specifically, to an improved low-frequency UWB SAR leaf cluster covert target fusion change detection method.

Background technique

In modern warfare, both sides pay more and more attention to the concealment of their own military targets, and at the same time improve the detection and reconnaissance capabilities of the enemy's hidden military targets. Therefore, the research on concealed target detection technology can provide important theoretical and technical support for the development of new battlefield reconnaissance/guided weapons and equipment in our country, and has important military significance. In addition, the geographical situation of my country's borders is complicated, and many areas are densely covered with jungles, which provides convenience for neighboring countries to deploy military targets near the borders and adjust their military defenses. Due to the shelter of the jungle, conventional radar systems cannot penetrate the jungle to effectively detect and reconnaissance the enemy. Therefore, there is an urgent need to develop advanced institutional radar systems and supporting detection systems to improve the detection and reconnaissance capabilities of hidden military targets.

Low-frequency ultra-wideband synthetic aperture radar (UWB SAR) has good foliage penetration detection performance and high-resolution imaging capability, and has become an important means of concealed target detection and reconnaissance. Due to the complex forest detection environment and the low-frequency UWB system, there are often many non-target strong scattering points (such as thick tree trunks) in the obtained UWB SAR images. These non-target speckled scattering points always have many false alarm points when performing change detection on low-frequency UWB SAR images of different phases, which increases the difficulty of detecting changes in leaf clusters.

At present, the low-frequency UWB SAR image target detection methods mainly include three types. The first type is based on the change detection method at the image pixel level. Through the comparison and analysis of the same pixel point in different phase images, it is observed whether the pixel point has changed. The common methods include the difference method and the ratio method. The implementation of the change detection method based on image pixel level is relatively simple, but the analysis of simple pixel gray value is greatly affected by noise, and the actual detection effect is often not good. The second type of method is a change detection method based on the image feature level. By extracting the features of the area around the pixel and comprehensively analyzing the extracted features, the detection of the change feature points is realized. Commonly used algorithms include the Edgeworth method and the gray level co-occurrence matrix method. The change detection method based on the image feature level uses image texture features, edge features and other information for comparison, which has strong stability and anti-interference ability, but it is relatively complicated to implement the change detection method based on the image pixel level. The third type of method is a change detection method based on the image target level. First, extract features and classify the single-temporal image, obtain the attribute information of the target point in the image, and then perform change detection on the classified image. Compared with the other two algorithms, the change detection method based on the image object level is a higher level change detection technology, and its research and application are extremely difficult, and it is still in its infancy.

Most of the above three methods are aimed at the change detection of a certain change detection information of the target point in the SAR image. It is difficult for a single change detection method to make full use of the change detection information in the SAR image, and it is easy to miss the detection target point. In recent years, target detection that combines multiple methods has attracted more and more attention from scholars. Improvement and fusion of various low-frequency UWB SAR leaf cluster concealed target detection methods can obtain better detection results. At present, the commonly used fusion change detection methods in the field of change detection include: Markov model method, and Laplacian eigenmap support vector description (LE-SVDD) method, etc. The fusion change detection method based on LE-SVDD can fuse multiple algorithms and has better detection performance than a single detection algorithm. However, due to the unreasonable sub-algorithm and detection threshold setting of the traditional fusion change detection method based on LE-SVDD, the fusion detection effect is not good, and it is difficult to achieve efficient and high-precision target detection in low-frequency UWB SAR images.

Contents of the invention

The invention provides an improved low-frequency UWB SAR leaf cluster covert target fusion change detection method, which can effectively suppress many non-target strong scattering points existing in the low-frequency UWB SAR image, and reduce the occurrence of false alarm points in the change detection process.

In order to achieve the above-mentioned technical effect, the technical scheme of the present invention is as follows:

An improved low-frequency UWB SAR leaf cluster covert target fusion change detection method, including the following steps:

S1: Preprocessing the detection image and the reference image;

S2: Calculate the inspection quantity image and detection threshold of the image segmentation difference method based on the improved clutter distribution model;

S3: Calculate the test quantity image and detection threshold of the one-dimensional Edgeworth method and the generalized Laguerre polynomial method;

S4: Input the three test quantity images and the detection threshold into the LE-SVDD classifier for training, and perform target and non-target discrimination on the test sample, which is the final change detection result.

Further, the process of step S1 includes:

Preprocess the detected image and the reference image, perform bidirectional relative radiation correction processing, and remove image gray value changes caused by system response and non-target changes in weather conditions in the image,

The specific process of step S1 is:

For each pixel in the observation area, a 15×15 pixel neighborhood is used to calculate the correlation coefficient value of the point between the reference image and the detection image. When the correlation coefficient is greater than 0.5, the point is set as a no-change point. For the detection image and reference image The no-change points in the entire image are respectively set to x ₁ , x ₂ ..., x _t and y ₁ , y ₂ ..., y _t ; the least square method is used for bidirectional linear estimation to obtain the corresponding linear coefficient and /> Use linear coefficients to assign weights to no-change points for weighted least squares estimation, and obtain corresponding linear coefficients /> and /> Then continue to assign weights until the linear coefficients stabilize, namely k _a , b _a , k _o and b _o , at this time:

In the formula, with /> are respectively the mean values of invariant points x ₁ , x ₂ ..., x _t and y ₁ , y ₂ ..., y _t ;

For all points of the detection image, a linear transformation is applied Transformation is performed to obtain the corrected detection image.

Further, the process of step S2 includes:

The image segmentation difference method based on the improved clutter distribution model is a pixel-level change detection method. The difference test image is obtained by subtracting the same point of the image in different time phases, and then the difference test image is segmented, and the probability density distribution of the non-target area of the segmented image is estimated, and a certain false alarm probability value is set. The corresponding detection threshold is obtained in the probability density distribution. If it is greater than the detection threshold, it is regarded as a change detection point, and thus the test image and detection threshold of the first sub-algorithm are obtained.

The concrete process of described step S2 is:

Let S ₁ and S ₂ be the intensity images of SAR observed at different times respectively, S ₁ is the reference image without the target, S ₂ is the detection image, and Z _dif is the difference inspection image composed of S ₁ and S ₂ :

First, carry out image segmentation to the reference image, adopt OTSU image segmentation algorithm, set the equivalent number of views of the reference image to be 1, the initial number of image divisions is N=2, segment the reference image, when the minimum value of the error square divided by the mean square of the various sub-regions after the reference image segmentation is less than the equivalent number of views, then stop segmenting, otherwise N=N+1, continue to segment the image until exiting the loop;

Next, estimate the clutter distribution of each area, assuming that the clutter distribution of each area of the difference test image is F _i (z _dif ), i∈N, set the false probability to 10 ^-8 , and the detection threshold corresponding to each area is The mathematical expression for the change detection of the new target in the image _S2 by using the image difference method is:

In the formula, z _dif , s ₁ , s ₂ are the gray value of the same pixel in the image Z _dif , S ₁ , S ₂ respectively, and T _dif is the detection threshold set according to the false alarm rate.

Further, the process of step S3 includes:

In the one-dimensional Edgeworth method, each pixel in the observation area is first centered, and the probability density function of the pixel gray value in its neighborhood is estimated based on the Edgeworth expansion based on the Gaussian probability density distribution model. The value probability density function is estimated based on the gamma probability density distribution model, and then the inspection quantity image is obtained through the K-L divergence; then, the image gray value statistics are performed on the inspection quantity image, and the curve of the statistical quantity is found to have a sharp drop on the left, and the inflection point in the steep drop curve is used as the inspection threshold; the least square method is used to fit the density curve during detection, and the horizontal axis value of the vertical axis equal to 0 in the fitted curve is approximated as the detection threshold, and the inspection quantity image and detection threshold of the two sub-algorithms are obtained.

The concrete process of described step S3 is:

Both the one-dimensional Edgeworth method and the generalized Laguerre polynomial method first estimate the probability density distribution, and then use the K-L divergence to calculate the probability density distribution inspection quantity image; for the corresponding point of the detection image and the reference image, it is necessary to calculate the mean value, variance, third-order cumulant, and fourth-order cumulant statistical information corresponding to the neighborhood of the point, and then substitute into the relevant formula derived from the K-L divergence to obtain the difference value of the probability distribution;

For the inspection quantity image obtained by the one-dimensional Edgeworth method and the generalized Laguerre polynomial method, since the number of non-target points in the detection image is far greater than the number of target points, grayscale statistics are carried out on the inspection quantity image. It is found that the non-target points are concentrated on the left side, and there is a sudden drop trend, and the target points are concentrated on the right side. The inflection point of the statistical curve is used as the detection threshold, and the method of fitting the statistical quantity curve of the inspection volume image is used on the right side to obtain the y=ax+b fitting curve. The x value corresponding to y=0 in the curve is detection threshold.

Further, the process of step S4 includes:

The SVDD classifier first records the change area detections of the change area and the no change area as target samples and outlier samples, and then uses the training sample set composed of pre-extracted target samples to train SVDD in order to construct a minimum hypersphere containing all training samples in the kernel feature space, and classify the change detection quantities in the observation scene based on this hypersphere; LE-SVDD sets the detection threshold and standard deviation of the three sub-algorithm test images as training samples based on SVDD. In the sub-algorithm, if there are one or two corresponding points of the test quantity image greater than the detection threshold and standard deviation, it is a test sample, and the training sample is input into the LE-SVDD classifier for training, and the target and non-target discrimination is performed on the test sample, which is the final change detection result.

The concrete process of described step S4 is:

For the detection image S ₁ and the reference image S ₂ , let the inspection image of the image segmentation difference method based on the improved clutter distribution model be I ₁ , the detection threshold be T ₁ , the inspection image obtained by the one-dimensional Edgeworth method and the generalized Laguerre polynomial method be I ₂ , I ₃ , and the detection threshold be T ₂ , T ₃ ;

Construct a sample data set as A={(i ₁ , i ₂ , i ₃ )|i ₁ ∈ I ₁ , i ₂ ∈ I ₂ , i ₃ ∈ I ₃ }, i ₁ , i ₂ , and i ₃ are all corresponding points in the test image I ₁ , I ₂ , I ₃ ;

Construct the data set for training as B={(i ₁ ,i ₂ ,i ₃ )|i ₁ -T ₁ >σ ₁ ,i ₂ -T ₂ >σ ₂ ,i ₃ -T ₃ >σ ₃ }, where σ ₁ , σ ₂ , and σ ₃ represent the standard deviations of the test image I ₁ , I ₂ , and I ₃ respectively;

The test data set is C={(i ₁ ,i ₂ ,i ₃ )||i ₁ -T ₁ |≤σ ₁ or |i ₂ -T ₂ |≤σ ₂ or |i ₃ -T| ₃ ≤σ ₃ };

The known non-target area is D={(i ₁ ,i ₂ ,i ₃ )|i ₁ -T ₁ <σ ₁ or i ₂ -T ₂ <σ ₂ or i ₃ -T<σ ₃ };

The sequential minimum optimization algorithm is used to train the training data set B, and then the trained LE-SVDD classifier is used to discriminate the test data set C, so as to obtain the information of the target point on the detection image, and then realize the high-efficiency and high-precision low-frequency UWB SAR leaf cluster covert target fusion change detection.

Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

The invention adopts the target fusion change detection technology based on LE-SVDD, which can effectively suppress many non-target strong scattering points existing in low-frequency UWB SAR images, and reduces the occurrence of false alarm points in the change detection process. While maintaining high target detection performance, the target detection efficiency is improved, thereby realizing high-efficiency and high-precision detection of low-frequency UWB SAR, and obtaining the position information of hidden targets. This method is suitable for detecting changes in military targets such as missile launchers and tanks hidden in the forest, and obtaining information about changes in the deployment of hidden targets in leaf clusters.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention;

Fig. 2 is the inspection image and reference image data adopted in the simulation experiment of the present invention;

Figure 3 is the detection result of the image segmentation difference method based on the improved clutter distribution model;

Fig. 4 is the detection result and statistical probability distribution figure of Edgeworth method and generalized Laguerre polynomial method;

Figure 5 shows the detection results of the improved low-frequency UWB SAR leaf cluster covert target fusion change detection method.

Detailed ways

The accompanying drawings are for illustrative purposes only and cannot be construed as limiting the patent;

In order to better illustrate this embodiment, some parts in the drawings will be omitted, enlarged or reduced, and do not represent the size of the actual product;

For those skilled in the art, it is understandable that some well-known structures and descriptions thereof may be omitted in the drawings.

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, an improved low-frequency UWB SAR leaf cluster covert target fusion change detection method includes the following steps:

S1: Preprocessing the detection image and the reference image;

S2: Calculate the inspection quantity image and detection threshold of the image segmentation difference method based on the improved clutter distribution model;

S3: Calculate the test quantity image and detection threshold of the one-dimensional Edgeworth method and the generalized Laguerre polynomial method;

S4: Input the three test quantity images and the detection threshold into the LE-SVDD classifier for training, and perform target and non-target discrimination on the test sample, which is the final change detection result.

Through the simulation experiment, the improved low-frequency UWB SAR leaf cluster covert target fusion change detection method of the present invention is verified, and the theoretical analysis and simulation experiment results prove the effectiveness of the present invention.

In the simulation experiment, the test image and reference image used are shown in Figure 2. The data set used for the detection image and the reference image is the imaging capture image of the Swedish CARABAS-II VHF SAR system in the VHF band (20-90MHz), and the imaging resolution is 2.5m×2.5m. 25 vehicle target points can be observed in the detection image, and there is no target point information in the reference image.

As shown in Figure 3, the image segmentation difference method based on the improved clutter distribution model is a pixel-level change detection method. The difference test image is obtained by subtracting the same point of the image in different phases, and then the difference test image is segmented, and the probability density distribution of the non-target area of the segmented image is estimated, and a certain false alarm probability value is set. The corresponding detection threshold is obtained in the probability density distribution. If it is greater than the detection threshold, it is regarded as a change detection point, and thus the test image and detection threshold of the first sub-algorithm are obtained.

The concrete process of step S2 is:

Let S ₁ and S ₂ be the intensity images of SAR observed at different times respectively, S ₁ is the reference image without the target, S ₂ is the detection image, and Z _dif is the difference inspection image composed of S ₁ and S ₂ :

First, carry out image segmentation to the reference image, adopt OTSU image segmentation algorithm, set the equivalent number of views of the reference image to be 1, the initial number of image divisions is N=2, segment the reference image, when the minimum value of the error square divided by the mean square of the various sub-regions after the reference image segmentation is less than the equivalent number of views, then stop segmenting, otherwise N=N+1, continue to segment the image until exiting the loop;

Next, estimate the clutter distribution of each area, assuming that the clutter distribution of each area of the difference test image is F _i (z _dif ), i∈N, set the false probability to 10 ^-8 , and the detection threshold corresponding to each area is The mathematical expression for the change detection of the new target in the image _S2 by using the image difference method is:

In the formula, z _dif , s ₁ , s ₂ are the gray value of the same pixel in the image Z _dif , S ₁ , S ₂ respectively, and T _dif is the detection threshold set according to the false alarm rate. It can be seen from Figure 3 that the image segmentation difference method based on the improved clutter distribution model has a good detection effect on the image, and can identify most of the target points, but there are still some target points missing.

As shown in Figure 4, in the one-dimensional Edgeworth method, each pixel in the observation area is first centered, and the probability density function of the pixel gray value in its neighborhood is estimated based on the Edgeworth expansion based on the Gaussian probability density distribution model. The pixel gray value probability density function is estimated based on the gamma probability density distribution model, and then the test image is obtained through the K-L divergence; then, the image gray value of the test image is counted, and the curve of the statistic is found to have a sharp drop on the left, and the inflection point in the drop curve is used as the test threshold; the least square method is used to fit the density curve during detection, and the horizontal axis value of the fitted curve is equal to 0.

The concrete process of step S3 is:

Both the one-dimensional Edgeworth method and the generalized Laguerre polynomial method first estimate the probability density distribution, and then use the K-L divergence to calculate the probability density distribution inspection quantity image; for the corresponding point of the detection image and the reference image, it is necessary to calculate the mean value, variance, third-order cumulant, and fourth-order cumulant statistical information corresponding to the neighborhood of the point, and then substitute into the relevant formula derived from the K-L divergence to obtain the difference value of the probability distribution;

For the inspection quantity image obtained by the one-dimensional Edgeworth method and the generalized Laguerre polynomial method, since the number of non-target points in the detection image is far greater than the number of target points, grayscale statistics are carried out on the inspection quantity image. It is found that the non-target points are concentrated on the left side, and there is a sudden drop trend, and the target points are concentrated on the right side. The inflection point of the statistical curve is used as the detection threshold, and the method of fitting the statistical quantity curve of the inspection volume image is used on the right side to obtain the y=ax+b fitting curve. The x value corresponding to y=0 in the curve is detection threshold. It can be seen from Figure 4 that by fitting the probability distribution curve and using the approximate inflection point in the curve as the detection threshold to detect the inspection image, almost all targets can be detected, with good detection performance, but there are some false alarm points in the detection results. Different from the image segmentation difference method based on the improved clutter distribution model, the Edgeworth method and the generalized Laguerre polynomial method analyze the differences in the statistical distribution characteristics of the image, but it is difficult to take into account the detection probability and the false alarm rate for the change detection of a single detection object.

As shown in Figure 5, the SVDD classifier first records the detection quantities of the change regions in the change region and the non-change region as target samples and outlier samples, and then uses the training sample set composed of pre-extracted target samples to train SVDD, in order to construct a minimum hypersphere containing all training samples in the kernel feature space, and classify the change detection quantities in the observation scene based on this hypersphere; on the basis of SVDD, LE-SVDD sets the detection threshold and standard deviation of the three sub-algorithms. In the three sub-algorithms, there are one or two corresponding points of the test quantity image greater than the detection threshold and standard deviation as the test class sample, and the training sample is input into the LE-SVDD classifier for training, and the target and non-target discrimination is performed on the test sample, which is the final change detection result.

The concrete process of step S4 is:

For the detection image S ₁ and the reference image S ₂ , let the inspection image of the image segmentation difference method based on the improved clutter distribution model be I ₁ , the detection threshold be T ₁ , the inspection image obtained by the one-dimensional Edgeworth method and the generalized Laguerre polynomial method be I ₂ , I ₃ , and the detection threshold be T ₂ , T ₃ ;

Construct a sample data set as A={(i ₁ , i ₂ , i ₃ )|i ₁ ∈ I ₁ , i ₂ ∈ I ₂ , i ₃ ∈ I ₃ }, i ₁ , i ₂ , and i ₃ are all corresponding points in the test image I ₁ , I ₂ , I ₃ ;

Construct the data set for training as B={(i ₁ ,i ₂ ,i ₃ )|i ₁ -T ₁ >σ ₁ ,i ₂ -T ₂ >σ ₂ ,i ₃ -T ₃ >σ ₃ }, where σ ₁ , σ ₂ , and σ ₃ represent the standard deviations of the test image I ₁ , I ₂ , and I ₃ respectively;

The test data set is C={(i ₁ ,i ₂ ,i ₃ )||i ₁ -T ₁ |≤σ ₁ or |i ₂ -T ₂ |≤σ ₂ or |i ₃ -T| ₃ ≤σ ₃ };

The known non-target area is D={(i ₁ ,i ₂ ,i ₃ )|i ₁ -T ₁ <σ ₁ or i ₂ -T ₂ <σ ₂ or i ₃ -T<σ ₃ };

The sequential minimum optimization algorithm is used to train the training data set B, and then the trained LE-SVDD classifier is used to discriminate the test data set C, so as to obtain the information of the target point on the detection image, and then realize the high-efficiency and high-precision low-frequency UWB SAR leaf cluster covert target fusion change detection. It can be seen from Fig. 5 that the present invention has a good detection effect on low-frequency UWB SAR images, and can detect the change information of SAR images in different phases. While maintaining a certain detection probability, the false alarm rate is reduced, and the detection speed is faster than the traditional LE-SVDD-based fusion change detection method of leaf cluster concealed targets. Therefore, the method of the present invention is an efficient and high-precision change detection method for low-frequency UWB SAR images.

The same or similar reference numerals correspond to the same or similar components;

The positional relationship described in the drawings is only for illustrative purposes and cannot be construed as a limitation to this patent;

Apparently, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the implementation of the present invention. For those of ordinary skill in the art, other changes or changes in different forms can be made on the basis of the above description. It is not necessary and impossible to exhaustively list all the implementation manners here. All modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (6)

1. An improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method is characterized by comprising the following steps:

s1: preprocessing a detection image and a reference image;

s2: obtaining a check quantity image and a detection threshold value based on an image segmentation difference method of an improved clutter distribution model;

the image segmentation difference method based on the improved clutter distribution model is a pixel-level change detection method, a difference value inspection amount image is obtained by subtracting the same point of images in different time phases, then the difference value inspection amount image is segmented, probability density distribution of a non-target area of the segmented image is estimated, a certain false alarm probability value is set, a corresponding detection threshold value is obtained in the probability density distribution, and if the probability density distribution is larger than the detection threshold value, the detection threshold value is regarded as a change detection point, so that the inspection amount image and the detection threshold value of a first sub-algorithm are obtained;

s3: obtaining an inspection amount image and a detection threshold value of a one-dimensional edge worth method and a generalized Laguerre polynomial method;

in the one-dimensional edge worth method, each pixel point of an observation area is taken as a center, an estimation based on a Gaussian probability density distribution model is carried out on a pixel gray value probability density function in the neighborhood of the observation area based on an edge worth expansion, and the difference of the probability density functions of each point between multi-time-phase images is analyzed based on a K-L divergence theory, so that a test quantity image related to the probability density difference is obtained; the generalized Laguerre polynomial method is based on a generalized Laguerre polynomial, the probability density function of the pixel gray value in the neighborhood of the generalized Laguerre polynomial is estimated based on a gamma probability density distribution model, and then a check-out image is obtained through K-L divergence; then, carrying out image gray value statistics on the inspection quantity image, finding out the trend of suddenly falling of a statistic curve on the left side, and adopting an inflection point in the suddenly falling curve as an inspection threshold; fitting a density curve by adopting a least square method during detection, and taking a horizontal axis value, equal to 0, of a vertical axis in the fitted curve as a detection threshold value to obtain a detection quantity image and a detection threshold value of two sub-algorithms;

s4: inputting three inspection quantity images and detection threshold values into an LE-SVDD classifier for training, and judging the target and non-target of a test sample, namely a final change detection result;

firstly, marking the detection amount of a change area and the detection amount of a change-free area as a target class sample and an outlier sample by an SVDD classifier, then training the SVDD by utilizing a training sample set formed by pre-extracted target class samples so as to construct a minimum hypersphere containing all training samples in a nuclear feature space, and classifying the detection amount of the change in an observation scene based on the hypersphere; based on SVDD, LE-SVDD sets detection threshold values and standard deviations, which are larger than corresponding detection quantity images, of all three sub-algorithm detection quantity images as training samples, wherein one or two detection quantity images corresponding to the detection threshold values and the standard deviations are larger than the detection threshold values and the standard deviations are test samples in the three sub-algorithms, inputs the training samples into an LE-SVDD classifier for training, and judges the target and the non-target of the test samples, namely the final change detection result.

2. The improved low frequency UWB SAR leaf cluster hidden target fusion change detection method of claim 1, wherein the process of step S1 comprises:

preprocessing the detection image and the reference image, carrying out bidirectional relative radiation correction processing, and removing image gray value change in the image caused by system response and weather condition non-target change factors.

3. The improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method according to claim 2, wherein the specific process of step S1 is:

for each pixel point of the observation area, adoptsThe pixel neighborhood of (2) calculates the correlation coefficient value of the point between the reference image and the detection image, when the correlation coefficient is larger than 0.5, the point is set as a no-change point, and the no-change point in the whole image in the detection image and the reference image is respectively set as ++>And->The method comprises the steps of carrying out a first treatment on the surface of the Carrying out bidirectional linear estimation by adopting a least square method to obtain a corresponding linear coefficient +.>、/>、/>And->The method comprises the steps of carrying out a first treatment on the surface of the Weighting the unchanged points by using the linear coefficients to perform weighted least square estimation to obtain corresponding linear coefficients +.>、/>、/>And->Continuing to assign weights until the linear coefficient is stable, namely +.>、/>、/>And->At this time:

，/>

in the method, in the process of the invention,and->Respectively is a constant point->And->Is the average value of (2);

for all points of the detected image, linear transformation is usedAnd performing transformation to obtain a corrected detection image.

4. The improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method according to claim 3, wherein the specific process of step S2 is:

is provided with、/>Observing the intensity images of the obtained SAR at different moments, respectively,/->For a reference image without target, +.>For detecting images +.>Is made of->And->The difference value test amount image is formed:

firstImage segmentation is carried out on the reference image, an OTSU image segmentation algorithm is adopted, the equivalent vision number of the reference image is set to be 1, and the initial segmentation number of the image is set to beDividing the reference image, stopping dividing when the minimum value of the square error divided by the square mean value of each sub-region after dividing the reference image is smaller than the equivalent vision number, otherwise +.>Continuing to segment the image until the loop is exited;

then, the clutter distribution of each region is estimated, and the clutter distribution of each region of the difference value test amount image is assumed to be，/>Setting the false probability as +.>The detection threshold value of each corresponding region is +.>，/>Image difference method is used for image->The mathematical expression for detecting the change of the newly appeared target is as follows:

in the method, in the process of the invention,、/>、/>respectively is image +.>、/>、/>Gray value of the same pixel point, +.>Is a detection threshold set according to the false alarm rate.

5. The improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method according to claim 4, wherein the specific process of step S3 is:

the one-dimensional edge worth method and the generalized Laguerre polynomial method are used for estimating probability density distribution firstly, and then calculating a probability density distribution check quantity image by adopting K-L divergence; for a corresponding point of the detection image and the reference image, calculating the statistical information of the mean value, the variance, the third-order accumulation amount and the fourth-order accumulation amount corresponding to the neighborhood of the point, substituting the statistical information into a related formula deduced by the K-L divergence, and solving a probability distribution difference value;

for the inspection quantity images obtained by a one-dimensional edge worth method and a generalized Laguerre polynomial method, as the number of non-target points in the detected image is far greater than the number of target points, the inspection quantity images are subjected to gray statistics, the non-target points are found to be concentrated on the left side and have a suddenly descending trend, the target points are concentrated on the right side, a statistic curve inflection point is adopted as a detection threshold, and the right side is fitted with the inspection quantity imagesIs to obtain the statistics curveFitting a curve, in which +.>Corresponding->The value is the detection threshold.

6. The improved low-frequency UWB SAR leaf cluster hidden target fusion change detection method according to claim 5, wherein the specific process of step S4 is:

for detected imagesAnd reference image->Let the test quantity image based on the image segmentation difference method of the improved clutter distribution model be +.>The detection threshold is->The inspection quantity images obtained by adopting a one-dimensional edge worth method and a generalized Laguerre polynomial method are respectively ++>、/>The detection threshold values are respectively->、/>；

Constructing a sample dataset as，/>、/>、/>Are all inspection volume images +.>、/>、/>Corresponding to the point in the first row;

constructing a training dataset asWherein->、/>、Respectively represent inspection quantity image +.>、/>、/>Standard deviation of (2);

the test data set is；

The non-target area is known as；

Utilization of training data sets using sequential minimum optimization algorithmsTraining and then using the trained LE-SVDD classifier to test data set +.>And judging to obtain information of a target point on the detected image, and further realizing high-efficiency high-precision low-frequency UWB SAR leaf cluster hidden target fusion change detection.