CN119540292A - High-precision video SAR moving target detection method based on information geometry theory - Google Patents

Abstract

The invention discloses a high-precision video SAR moving target detection method based on an information geometry theory, which belongs to the technical field of SAR moving target shadow detection. The high-precision video SAR moving target detection method based on the information geometric theory comprises the following steps of carrying out shadow enhancement and background suppression on a video SAR image sequence, extracting an ROI region based on the information geometric theory, and effectively removing false ROI regions through false detection suppression. The high-precision video SAR moving target detection method based on the information geometric theory improves the detection probability of the shadow of the video SAR moving target on the premise of ensuring the detection precision.

Description

High-precision video SAR moving target detection method based on information geometric theory

Technical Field

The invention relates to the technical field of SAR moving target shadow detection, in particular to a high-precision video SAR moving target detection method based on an information geometric theory.

Background

The video SAR has the capability of continuously monitoring the change of a target area, and has wide application in the fields of military decision making, smart cities, traffic supervision and the like. In video SAR systems, the doppler modulation of moving object echoes is very sensitive to the movement of the object due to the high carrier frequency of the radar. Even small movements can cause large shifts and defocusing of moving object imaging in the SAR image, but shadows left by the moving object can reflect the actual position and state information of the moving object in the SAR image sequence. Therefore, shadows of moving objects are more easily observed in a video SAR image sequence, and detection with moving object shadows is a more direct and efficient means than in conventional SAR systems.

Currently, in the conventional video SAR moving target shadow detection technology, there are three main types of methods, namely a moving target detection method based on difference, a constant false alarm detection (CFAR) based device and a threshold segmentation based method.

According to the detection method based on the difference, firstly, a background image is estimated through a plurality of continuous video SAR images, then the background image and the current image to be detected are subjected to difference operation to obtain a binary image of the foreground, then the motion information of the previous frame and the motion information of the next frame are extracted through an inter-frame difference algorithm, and the binary image obtained through the difference between the motion information and the background are subjected to AND operation, and finally, the previous step of result is subjected to morphological processing to extract a shadow region of a motion target. Although the differential-based detection method is effective, the method has some limitations that firstly, the background modeling algorithm has higher requirements, the background is subject to hysteresis due to the fact that a plurality of previous frame estimation is used, and secondly, the inter-frame differential algorithm only compares the difference between two frames and ignores the information of the overlapped part of the two frames, so that the phenomena of hollowness and bilateral in a shadow area of a moving object are easy to occur, and the condition of missed detection or false alarm is caused.

The CFAR-based detection method can effectively avoid the limitation, the CFAR can select a proper area in the background noise of the video SAR image, and a proper detection threshold is determined according to the noise statistical characteristics in the area. When the target area exceeds this threshold, the target is considered detected. However, in a multi-target environment, multiple target signals may be included in the reference area, which may reduce the detection performance of the method, while in a non-uniform noise environment, the performance of the CFAR may be affected.

The detection method based on threshold segmentation can binarize the video SAR image by using an image threshold segmentation method (such as an OTSU method, a minimum error method and the like) to obtain a moving target shadow region in the image. However, the threshold segmentation method is based on the histogram information of the image, and ignores the spatial information and the edge information of the video SAR image, so that the problems of missed detection and over-high false alarm are easy to occur.

Disclosure of Invention

In order to solve the problems, the invention aims to provide a high-precision video SAR moving target detection method based on an information geometric theory, solve the technical problems of the defects of the traditional video SAR moving target shadow detection method, and improve the detection probability of the video SAR moving target shadow on the premise of ensuring the detection precision.

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

The invention provides a high-precision video SAR moving target detection method based on an information geometric theory, which comprises the following steps:

s1, performing shadow enhancement and background suppression on a video SAR image sequence;

s2, extracting an ROI region based on an information geometric theory;

s3, effectively removing false ROI areas through error detection inhibition.

Further, the step S1 includes the following steps:

s11, calculating an initial subspace { U, V } of the video SAR image sequence by using a principal component analysis algorithm;

S12, setting the number of mixed gaussians, and modeling a data matrix X formed by a video SAR image sequence based on a low-rank representation model of the mixed gaussians, namely:

Wherein,Representing a set of all gaussian component weights,Representing a set of all gaussian component variances;

S13, updating a low-rank matrix U-V^T by using an online subspace learning technology, and subtracting the low-rank matrix U-V^T from an original video SAR image to obtain a foreground image containing video SAR moving target shadows;

s14, removing strong scattered point object imaging contained in the foreground image by using an alternate direction multiplier algorithm to obtain a video SAR image sequence subjected to shadow enhancement and background suppression, namely an enhanced image.

Further, the step S2 includes the following steps:

S21, forming a neighborhood system by all pixels in the neighborhood of the enhanced image at the coordinate pixel, and constructing statistical distribution describing the point;

s22, calculating geometrical distances in a JS divergence description Gaussian statistical manifold for mode analysis, and capturing local information features in an image;

S23, traversing each pixel of the image to obtain a corresponding JS divergence matrix of the enhanced image, and calculating joint probability distribution JPDF;

s24, performing threshold selection by using a minimum cross entropy algorithm, and extracting the ROI region of the enhanced image.

Further, the step S21 includes the following steps:

S211, forming a neighborhood system by all pixels in N x N neighborhood of the enhanced image at the (x, y) pixels, calculating the mean mu and variance sigma of all pixels in the neighborhood system, and describing the statistical distribution of the neighborhood system by using Gaussian distribution:

s212, taking the gray intensity value at the current (x, y) pixel as a mean value mu ', giving a minimum variance sigma', and constructing a statistical distribution describing the point:

further, the step S22 includes the following steps:

s221, a statistical model S= { P (x|theta) |theta epsilon theta } is formed by a probability distribution family formed by parameters theta= (mu, sigma)^T, and S forms a micro Gaussian statistical manifold under a certain topological structure;

S222, calculating JS divergence, describing geometric distances among different probability distributions in the Gaussian statistical manifold by adopting the JS divergence, and analyzing modes;

S223, normalizing the value of the JS divergence, and explaining the geometric relationship between the pixel gray value of the current position and the neighborhood system distribution of the pixel gray value on a Gaussian statistical manifold to help capture local information features in an image.

Further, the JS divergence calculation formula in S222 is as follows:

Wherein,Is the average of two distributions, KL (pq) is defined as:

Further, the step S23 includes the following steps:

s231, traversing each pixel of the image to obtain a corresponding JS divergence matrix of the enhanced image;

S232, setting n_ij as the joint occurrence number between the gray value and JS scattering, and dividing each element in n_ij by the total pixel number of the enhanced image to obtain normalized joint probability distribution JPDF.

Further, the step S3 includes the following steps:

s31, calculating a gradient image of the current frame, and eliminating an error ROI region by using the gradient image;

S32, performing multi-frame false alarm suppression by using a data association algorithm.

Further, the step S31 includes the following steps:

S311, detecting the edge of the image by adopting a Sobel operator, and calculating the gray level relative change of the original video SAR image to obtain a gradient image of the current original SAR image;

S312, introducing the gradient image, calculating the superposition area of the gradient image and the ROI, and eliminating the ROI with the superposition area larger than the threshold value.

Further, the step S32 includes the following steps:

s321, extracting the mass center position of each ROI area and the corresponding time frame number as a feature vector;

s322, setting a space distance threshold T_I and a time threshold T_T, and associating detection points in a time threshold frame through the two thresholds to identify a possible real target;

S323, the detection point with stronger consistency in a plurality of frames is regarded as an effective target, and the detection point which accords with certain association times and the spatial variation of the centroid can be reserved.

By adopting the technical scheme, the invention has the following advantages:

1. The invention provides a high-precision video SAR moving target detection method based on an information geometric theory, which enhances the representation of the shadow of a moving target and simultaneously effectively inhibits strong scattered point objects and low scattering areas in the background by carrying out the operations of shadow enhancement and background inhibition on a video SAR image sequence.

2. The invention provides a high-precision video SAR moving target detection method based on an information geometric theory, which calculates the geometric distance between the current pixel position and the neighborhood system distribution thereof based on the information geometric theory, reveals the change of a local structure, forms joint probability distribution with gray value distribution, completes the extraction of an ROI region by utilizing minimum cross entropy, and greatly improves the detection performance of moving target shadows.

3. The invention provides a high-precision video SAR moving target detection method based on an information geometric theory, which utilizes dynamic information between edge information and multiple frames to effectively remove false ROI areas and improve the detection precision of moving target shadows.

Drawings

FIG. 1 is a flow chart of a high-precision video SAR moving target detection method based on information geometry theory;

FIG. 2 is a flow chart of ROI region extraction according to the present invention;

FIG. 3 (a) is an original video SAR image of the present invention;

FIG. 3 (b) is an enhanced image of the present invention;

FIG. 4 is an extracted ROI area of the present invention;

Fig. 5 is a detection result of a moving object per frame of the present invention.

Detailed Description

In the following detailed description of the embodiments of the present invention, reference is made to the accompanying drawings, in which it is to be noted that, in this document, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Fig. 1 shows a flow chart of a high-precision video SAR moving target detection method based on the information geometric theory, and the high-precision video SAR moving target detection method based on the information geometric theory is wholly divided into three parts, wherein the method comprises the following steps:

s1, performing shadow enhancement and background suppression on a video SAR image sequence;

wherein, S1 comprises the following specific steps:

s11, calculating an initial subspace { U, V } of the video SAR image sequence by using a principal component analysis algorithm;

S12, setting the number of mixed gaussians, and modeling a data matrix X formed by a video SAR image sequence based on a low-rank representation model of the mixed gaussians, namely:

Wherein,Representing a set of all gaussian component weights,Representing a set of all gaussian component variances;

S13, updating a low-rank matrix U-V^T by using an online subspace learning technology, and subtracting the low-rank matrix U-V^T from an original video SAR image to obtain a foreground image containing video SAR moving target shadows;

S14, removing strong scattered point object imaging contained in the foreground image by using an alternate direction multiplier algorithm to obtain a video SAR image sequence subjected to shadow enhancement and background suppression, namely an enhanced image for short.

S2, extracting an ROI region based on an information geometric theory;

Wherein, the following steps are specifically shown in fig. 2:

S21, forming a neighborhood system by all pixels in the neighborhood of the enhanced image at the coordinate pixel, and constructing statistical distribution describing the point;

Wherein, S21 comprises the following specific steps:

S211, forming a neighborhood system by all pixels in N x N neighborhood of the enhanced image at the (x, y) pixels, calculating the mean mu and variance sigma of all pixels in the neighborhood system, and describing the statistical distribution of the neighborhood system by using Gaussian distribution:

s212, taking the gray intensity value at the current (x, y) pixel as a mean value mu ', giving a minimum variance sigma', and constructing a statistical distribution describing the point:

s22, calculating geometrical distances in a JS divergence description Gaussian statistical manifold for mode analysis, and capturing local information features in an image;

wherein, S22 comprises the following specific steps:

S221, a probability distribution family formed by parameters theta= (mu, sigma)^T can form a statistical model S= { P (x|theta) |theta ∈theta }, S forms a micro Gaussian statistical manifold under a certain topological structure, and compared with Euclidean space and Euclidean distance measurement, the statistical manifold and the measurement based on the geodesic distance can reflect the geometric position and the similarity of each point;

S222, calculating by adopting Jensen-Shannon divergence (JS divergence), describing geometric distances among different probability distributions in the Gaussian statistical manifold by adopting the JS divergence, and analyzing modes;

Wherein,Is the average of two distributions, KL (p||q) is Kullback-Leibler divergence (KL divergence), defined as:

S223, normalizing the value of JS divergence to 0-255, on Gaussian statistical manifold, JS divergence provides a non-Euclidean 'distance' measure, the geometrical relationship between the pixel gray value of the current position and the neighborhood system distribution can be explained on manifold, the change of local structure can be revealed, and the capturing of local information features in an image can be facilitated, wherein the features are 'target areas' different from the background.

S23, traversing each pixel of the image to obtain a corresponding JS divergence matrix of the enhanced image, and calculating joint probability distribution JPDF;

S23 comprises the following specific steps:

s231, traversing each pixel of the image to obtain a corresponding JS divergence matrix of the enhanced image;

S232, setting n_ij as the joint occurrence number between the gray value and JS scattering, and dividing each element in n_ij by the total pixel number of the enhanced image to obtain normalized joint probability distribution JPDF.

S24, based on the obtained joint probability distribution JPDF, calculating JPDF = { jpdf_ij, i, j=0, 1, & gt, and extracting the ROI area of the enhanced image by using a minimum cross entropy algorithm, wherein the optimal gray value threshold T_G of L-1 and the JS divergence threshold T_JS.

S3, false ROI areas are effectively removed through error detection suppression (road edge information assistance and inter-frame association).

S3 comprises the following specific steps:

s31, calculating a gradient image of the current frame, and eliminating an error ROI region by using the gradient image;

wherein, S31 comprises the following specific steps:

s311, detecting the edge of the image by adopting a Sobel operator, and calculating the gray level relative change of the original video SAR image to obtain a gradient image of the current original SAR image;

S312, introducing the gradient image, calculating the superposition area of the gradient image and the ROI area, and eliminating the ROI area with the superposition area larger than the threshold T_A.

S32, most of the false ROI areas existing at the edge are eliminated, but a small part of the false ROI areas exist, so that the false alarm suppression of multiple frames is performed by using a data association algorithm.

Wherein, S32 comprises the following specific steps:

s321, extracting the mass center position of each ROI area and the corresponding time frame number as a feature vector;

s322, setting a space distance threshold T_I and a time threshold T_T, and associating detection points in a time threshold frame through the two thresholds to identify a possible real target;

S323, the detection point with stronger consistency in a plurality of frames is regarded as an effective target, and the detection point which accords with certain association times and the spatial variation of the centroid can be reserved.

Finally, high-precision moving target detection of the video SAR can be realized through the three major parts.

In a specific embodiment, the implementation steps are as follows:

S1, performing shadow enhancement and background suppression on an original video SAR image by adopting a low-rank representation model based on mixed Gaussian to obtain an enhanced image sequence, in the example, selecting a 56 th frame video SAR image and an enhanced image as examples are shown in fig. 3 (a) and 3 (b), and comparing the fig. 3 (a) and 3 (b) to show that the shadow of a moving target of the enhanced image is more obvious and the contrast of the moving target shadow with the surrounding is higher, and simultaneously suppressing the background of the original video SAR image so that the extraction operation of a subsequent ROI area of the original video SAR image is not interfered by the background.

S2, setting the neighborhood size of 20 x 20, extracting a JS divergence characteristic matrix of an image by utilizing an information geometric theory, combining the JS divergence characteristic matrix with an image gray value, calculating joint probability distribution JPDF of the image, calculating an optimal gray value threshold T_G and a JS divergence threshold T_JS by adopting a minimum cross entropy algorithm, and simultaneously, setting the pixel point positions of the threshold T_G and the JS divergence threshold T_JS smaller than the threshold pixel point positions as the ROI region. And carrying out connected domain statistics on the ROI area, and deleting the connected domains of more than 500 pixels and less than 100 pixels. And recording centroid position information and frame number information of the connected domain, and entering the next step. The overall flow of this step is shown in fig. 2, and an exemplary diagram of the extracted ROI area is shown in fig. 4 in particular, and it can be seen from fig. 4 that almost all the moving object shadow areas are detected.

S3, calculating the gradient amplitude of the current original video SAR image by utilizing a Sobel operator, and generating a gradient image with the gradient amplitude of more than 50. And calculating the superposition area of the gradient image and the ROI area, wherein the superposition area is more than 50% of the area of the ROI area of the gradient image and the ROI area is the error ROI area, and eliminating the error ROI area. And then extracting the centroid position and the time frame number of the ROI area of each frame, and calculating the Euclidean distance between the detection result of the current frame and the detection result of the next frame by frame. If a target is found to be less than the spatial distance threshold (T_I =30) from the target in the next frame, then both targets are considered to belong to the same track and marked as the same group, if not, then found in the next frame until the limit of the temporal threshold (T_T =5) is exceeded. Finally, counting the association times of each target, and detecting the change between the first centroid position and the last centroid position. If the movement distance is large and the association number is 8 or more, the target is considered to be a valid target. The final detection result of the moving object of each frame is shown in fig. 5, wherein the red frame is the detection frame of the present invention, and the green frame is the real label frame.

Finally, it is pointed out that while the invention has been described with reference to a specific embodiment thereof, it will be understood by those skilled in the art that the above embodiments are provided for illustration only and not as a definition of the limits of the invention, and various equivalent changes or substitutions may be made without departing from the spirit of the invention, therefore, all changes and modifications to the above embodiments shall fall within the scope of the appended claims.

Claims (10)

Translated fromChinese

1.基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，包括以下步骤：1. A high-precision video SAR moving target detection method based on information geometry theory, characterized in that it comprises the following steps:S1.对视频SAR图像序列进行阴影增强与背景抑制；S1. Perform shadow enhancement and background suppression on video SAR image sequences;S2.基于信息几何理论的ROI区域提取；S2. ROI region extraction based on information geometry theory;S3.通过错检抑制，有效去除虚假ROI区域。S3. Effectively remove false ROI areas through false detection suppression.2.根据权利要求1所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S1包括以下步骤：2. The high-precision video SAR moving target detection method based on information geometry theory according to claim 1, characterized in that said S1 comprises the following steps:S11.利用主成分分析算法计算视频SAR图像序列的初始子空间{U,V}；S11. Calculate the initial subspace {U, V} of the video SAR image sequence using the principal component analysis algorithm;S12.设置混合高斯个数，并对视频SAR图像序列所组成的数据矩阵X进行基于混合高斯的低秩表示模型建模，即：S12. Set the number of mixed Gaussians, and model the data matrix X composed of the video SAR image sequence based on a mixed Gaussian low-rank representation model, that is:其中，表示所有高斯分量权重的集合，表示所有高斯分量方差的集合；in, represents the set of all Gaussian component weights, represents the set of variances of all Gaussian components;S13.利用在线子空间学习技术更新低秩矩阵U*V^T，与原始的视频SAR图像相减得到包含视频SAR运动目标阴影的前景图像；S13. Update the low-rank matrix U*V^T using online subspace learning technology, and subtract it from the original video SAR image to obtain a foreground image containing the shadow of the video SAR moving target;S14.利用交替方向乘子算法去除前景图像中包含的强散点物体成像，得到经过阴影增强与背景抑制的视频SAR图像序列，即增强图像。S14. Using the alternating direction multiplier algorithm to remove the strong scattered object imaging contained in the foreground image, a video SAR image sequence with shadow enhancement and background suppression, namely, an enhanced image, is obtained.3.根据权利要求1所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S2包括以下步骤：3. The high-precision video SAR moving target detection method based on information geometry theory according to claim 1, characterized in that said S2 comprises the following steps:S21.将增强图像在坐标像素处邻域内的所有像素构成一个邻域系统，构建描述该点的统计分布；S21. All pixels in the neighborhood of the coordinate pixel of the enhanced image form a neighborhood system, and construct a statistical distribution describing the point;S22.计算JS散度描述高斯统计流形中的几何距离进行模式的分析，捕捉图像中的局部信息特征；S22. Calculate the JS divergence to describe the geometric distance in the Gaussian statistical manifold to analyze the pattern and capture the local information features in the image;S23.遍历图像的每个像素，得到增强图像相应的JS散度矩阵，计算联合概率分布JPDF；S23. Traverse each pixel of the image, obtain the JS divergence matrix corresponding to the enhanced image, and calculate the joint probability distribution JPDF;S24.利用最小交叉熵算法进行阈值选取，提取增强图像的ROI区域。S24. Use the minimum cross entropy algorithm to select the threshold and extract the ROI area of the enhanced image.4.根据权利要求3所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S21包括以下步骤：4. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3, characterized in that said S21 comprises the following steps:S211.将增强图像在(x,y)像素处N*N邻域内的所有像素构成一个邻域系统，计算该邻域系统内所有像素的均值μ和方差σ，利用高斯分布描述该邻域系统的统计分布：S211. All pixels in the N*N neighborhood of the enhanced image at the (x, y) pixel form a neighborhood system, calculate the mean μ and variance σ of all pixels in the neighborhood system, and use Gaussian distribution to describe the statistical distribution of the neighborhood system:S212.以当前(x,y)像素处的灰度强度值为均值μ′，给定一个极小的方差σ′，构建描述该点的统计分布：S212. Taking the gray intensity value at the current (x, y) pixel as the mean μ′ and a very small variance σ′, construct a statistical distribution describing the point:5.根据权利要求3所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S22包括以下步骤：5. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3, characterized in that said S22 comprises the following steps:S221.由参数θ＝(μ,σ)^T构成的概率分布族组成一个统计模型S＝{P(x|θ)|θ∈Θ}，S在一定的拓扑结构下构成一个可微的高斯统计流形；S221. The probability distribution family composed of parameters θ = (μ, σ)^T forms a statistical model S = {P(x|θ)|θ∈Θ}, and S forms a differentiable Gaussian statistical manifold under a certain topological structure;S222.计算JS散度，采用JS散度来描述不同概率分布之间在该高斯统计流形中的几何距离，进行模式的分析；S222. Calculate JS divergence, use JS divergence to describe the geometric distance between different probability distributions in the Gaussian statistical manifold, and perform pattern analysis;S223.将JS散度的数值归一化，在高斯统计流形上，JS散度解释当前位置的像素灰度值与其邻域系统分布之间的几何关系，帮助捕捉图像中的局部信息特征。S223. Normalize the numerical value of JS divergence. On the Gaussian statistical manifold, JS divergence explains the geometric relationship between the pixel grayscale value at the current position and the distribution of its neighborhood system, helping to capture the local information features in the image.6.根据权利要求5所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S222中的JS散度计算公式如下：6. The high-precision video SAR moving target detection method based on information geometry theory according to claim 5, characterized in that the JS divergence calculation formula in S222 is as follows:其中，是两个分布的平均分布，KL(p||q)定义为：in, is the average distribution of the two distributions, and KL(p||q) is defined as:7.根据权利要求3所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S23包括以下步骤：7. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3, characterized in that said S23 comprises the following steps:S231.遍历图像的每个像素，得到该增强图像相应的JS散度矩阵；S231. Traverse each pixel of the image to obtain the JS divergence matrix corresponding to the enhanced image;S232.设n_ij为灰度值和JS散度之间的联合出现次数，将n_ij中的每个元素除以增强图像的总像素数得到归一化后的联合概率分布JPDF。S232. Let n_ij be the number of joint occurrences between the gray value and the JS divergence, and divide each element in n_ij by the total number of pixels of the enhanced image to obtain the normalized joint probability distribution JPDF.8.根据权利要求3所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S3包括以下步骤：8. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3, characterized in that said S3 comprises the following steps:S31.计算当前帧的梯度图像，利用所述梯度图像进行错误ROI区域的剔除；S31. Calculate the gradient image of the current frame, and use the gradient image to remove the erroneous ROI area;S32.利用数据关联算法进行多帧的虚警抑制。S32. Use data association algorithm to suppress false alarms of multiple frames.9.根据权利要求8所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S31包括以下步骤：9. The high-precision video SAR moving target detection method based on information geometry theory according to claim 8, characterized in that said S31 comprises the following steps:S311.采用Sobel算子对图像边缘处进行检测，计算原始视频SAR图像的灰度级相对变化，得到当前原始SAR图像的梯度图像；S311. Use the Sobel operator to detect the edge of the image, calculate the relative change of the gray level of the original video SAR image, and obtain the gradient image of the current original SAR image;S312.引入该梯度图像，计算该梯度图像与ROI区域的重合面积，对重合面积大于阈值的ROI区域进行剔除。S312. Import the gradient image, calculate the overlap area between the gradient image and the ROI region, and remove the ROI region whose overlap area is greater than a threshold.10.根据权利要求8所述的基于信息几何理论的高精度视频SAR运动目标检测方法，其特征在于，所述S32包括以下步骤：10. The high-precision video SAR moving target detection method based on information geometry theory according to claim 8, characterized in that said S32 comprises the following steps:S321.提取每个ROI区域的质心位置以及对应的时间帧编号，作为特征向量；S321. Extract the centroid position of each ROI area and the corresponding time frame number as a feature vector;S322.设定一个空间距离阈值T_I和时间阈值T_T，通过这两个阈值将时间阈值帧内的检测点进行关联，识别出可能的真实目标；S322. Set a spatial distance threshold_TI and a time threshold_TT , and associate the detection points in the time threshold frame through these two thresholds to identify possible real targets;S323.在多帧中一致性较强的检测点被认为是有效目标，符合一定的关联次数与质心的空间变化的检测点才能被保留。S323. Detection points with strong consistency in multiple frames are considered to be valid targets, and only detection points that meet a certain number of associations and spatial changes in the center of mass can be retained.