CN110728695A - Video SAR Moving Target Detection Method Based on Image Region Accumulation - Google Patents

Abstract

Translated fromChinese

本发明公开了一种基于动态阴影的视频SAR运动目标检测方法，主要解决现有视频SAR运动目标检测不稳健的问题。其实现方案为：1)设计积累窗并对视频SAR图像序列进行区域积累；2)确定积累门限并对积累结果进行图像重整；3)对重整后的图像二值分割；4)对二值分割后的图像进行连通域大小统计并保留大小在0.4倍目标所占像素点总数与2倍目标所占像素点总数范围内的连通域；5)对连通域处理后的图像进行帧间相关处理去除非目标阴影，完成运动目标检测。本发明通过对视频SAR图像进行区域化的多帧联合检测，有效抑制了虚警和漏警概率并提高了检测性能，可用于视频SAR运动目标的跟踪和实时探测。

The invention discloses a video SAR moving target detection method based on dynamic shadows, which mainly solves the problem that the existing video SAR moving target detection is not robust. The implementation scheme is: 1) design an accumulation window and perform regional accumulation on the video SAR image sequence; 2) determine the accumulation threshold and perform image reorganization on the accumulation result; 3) binary segmentation of the reshaped image; 4) two The size of the connected domain is counted on the image after value segmentation, and the connected domain whose size is within the range of 0.4 times the total number of pixels occupied by the target and 2 times the total number of pixels occupied by the target is retained; 5) Inter-frame correlation is performed on the image after the connected domain processing. Process to remove non-target shadows and complete moving target detection. The invention effectively suppresses the probability of false alarms and missed alarms and improves the detection performance by performing regionalized multi-frame joint detection on video SAR images, and can be used for tracking and real-time detection of video SAR moving targets.

Description

Translated fromChinese

基于图像区域积累的视频SAR运动目标检测方法Video SAR Moving Target Detection Method Based on Image Region Accumulation

技术领域technical field

本发明属于数据处理技术领域，特别涉及一种SAR运动目标检测方法，可用于视频SAR运动目标的跟踪和实时探测。The invention belongs to the technical field of data processing, in particular to a SAR moving target detection method, which can be used for tracking and real-time detection of video SAR moving targets.

背景技术Background technique

视频SAR成像雷达能够全天时、全天候、高精度地对地面目标区域进行实时成像，具有成像帧率高、分辨率高的特点，能有效克服红外/可见光传感器易受天气条件和战场环境影响的弱点，同时也能克服常规SAR系统帧率低、动目标检测跟踪困难等缺陷。视频SAR成像时，运动目标图像会发生散焦和移位，但会在其真实位置留下阴影，并且成像阴影会在视频成像模式下所获得的图像序列中发生移动。因此，基于动态阴影的数据处理技术从原理上可以实现对运动目标的检测和跟踪。然而，由于SAR使用相干电磁波进行成像，导致SAR图像中存在大量的相干斑噪声，利用阴影进行目标检测时将产生大量虚警，因此对视频SAR中的运动目标进行稳健的检测具有重要意义。Video SAR imaging radar can perform real-time imaging of ground target areas in all-day, all-weather and high-precision. It has the characteristics of high imaging frame rate and high resolution. At the same time, it can overcome the shortcomings of conventional SAR systems such as low frame rate and difficulty in moving target detection and tracking. During video SAR imaging, moving target images are defocused and shifted, but shadows are left at their true locations, and the imaging shadows move in the image sequence acquired in video imaging mode. Therefore, the data processing technology based on dynamic shadow can realize the detection and tracking of moving objects in principle. However, since SAR uses coherent electromagnetic waves for imaging, there is a large amount of speckle noise in SAR images, and a large number of false alarms will be generated when using shadows for target detection. Therefore, robust detection of moving targets in video SAR is of great significance.

现有的基于视频SAR中动态阴影的运动目标检测方法大多从图像配准的角度出发，通过提取静止背景的图像序列并与场景视频SAR图像序列相减，得到图像序列中的变化信息。该类方法是基于单像素点的操作，存在着背景提取不准确、虚警率过高的问题。近年来，越来越多的检测方法被应用在视频SAR的运动目标检测中，基于视频SAR图像帧的动目标检测技术也因此得到了发展。Most of the existing moving target detection methods based on dynamic shadows in video SAR start from the perspective of image registration, and obtain the change information in the image sequence by extracting the image sequence of the still background and subtracting it from the scene video SAR image sequence. This kind of method is based on the operation of single pixel point, and has the problems of inaccurate background extraction and high false alarm rate. In recent years, more and more detection methods have been used in the detection of moving objects in video SAR, and the technology of moving object detection based on video SAR image frames has also been developed.

Doerry A、Miller J、Bishop E等人在论文“Shadow Probability of Detectionand False Alarm for Median-Filtered SAR Imagery”中通过中值滤波方法抑制了相干斑噪声并提高了SAR图像中阴影区域的单点检测性能。Y Zhang、X Mao等人在论文“A novelapproach to moving targets shadow detection in videosar imagery sequence”中则针对Sandia实验室公布的视频数据通过图像处理的方式提取动态阴影从而达到动目标检测的目的。这两种方法均是基于图像像素的操作，由于没有考虑运动目标阴影的像素间关系，以及帧间的阴影时序变化关系，因而易出现虚警、漏警过高的问题，造成对低速目标以及高机动目标点的检测性能恶化。In the paper "Shadow Probability of Detection and False Alarm for Median-Filtered SAR Imagery" by Doerry A, Miller J, Bishop E et al., median filtering method suppresses speckle noise and improves single-point detection performance of shadow areas in SAR images . In the paper "A novelapproach to moving targets shadow detection in videosar imagery sequence" by Y Zhang, X Mao and others, they extract dynamic shadows through image processing for the video data published by Sandia's laboratory to achieve the purpose of moving target detection. These two methods are based on the operation of image pixels. Since the relationship between the pixels of the shadow of the moving target and the time sequence change of the shadow between frames are not considered, the problems of false alarms and missed alarms are prone to occur, resulting in low-speed targets and low-speed targets. The detection performance of high maneuvering target points deteriorates.

发明内容SUMMARY OF THE INVENTION

本发明的目的在于针对上述现有技术的不足，提出一种基于图像区域积累的视频SAR运动目标检测方法，以减小视频SAR中由帧间阴影时序变化带来的虚警和漏警影响，提高动目标的检测概率。The purpose of the present invention is to aim at the above-mentioned deficiencies of the prior art, and propose a video SAR moving target detection method based on image area accumulation, so as to reduce the influence of false alarms and missed alarms caused by inter-frame shadow timing changes in video SAR, Improve the detection probability of moving targets.

实现本发明目的的技术方案包括如下步骤：The technical scheme that realizes the object of the present invention comprises the following steps:

(1)对视频SAR图像序列进行区域积累:(1) Regional accumulation of video SAR image sequences:

1a)设置积累窗，根据视频SAR图像序列的分辨率和运动目标阴影在图像中的尺寸确定积累窗大小为N*N，其中N为奇数，其中*表示点乘；1a) the accumulation window is set, and the size of the accumulation window is determined to be N*N according to the resolution of the video SAR image sequence and the size of the shadow of the moving target in the image, where N is an odd number, and * represents a dot product;

1b)使用滑动积累窗的方法进行图像区域积累，积累窗中心从第一个图像像素开始逐点移动，每个对应像素点的值由对该点周围的积累窗内覆盖的图像强度值求和确定；1b) Use the method of sliding accumulation window to accumulate the image area, the center of the accumulation window moves point by point from the first image pixel, and the value of each corresponding pixel is summed by the image intensity values covered in the accumulation window around the point. Sure;

1c)重复1b)直至积累窗中心完全遍历视频SAR图像序列，生成积累后的图像序列；1c) Repeat 1b) until the center of the accumulation window completely traverses the video SAR image sequence to generate the accumulated image sequence;

(2)确定积累门限S_T并对积累结果进行图像重整:(2) Determine the accumulation threshold S_T and perform image reorganization on the accumulation result:

2a)根据均匀背景概率密度函数f_B(s)、系统噪声水平λ和对虚警率P_fa的要求,利用下式确定积累门限S_T，即：2a) According to the uniform background probability density function f_B (s), the system noise level λ and the requirements for the false alarm rate P_fa , use the following formula to determine the accumulation threshold S_T , namely:

2b)利用积累门限S_T对1c)中生成的积累后图像序列进行统一重整，即设置与1a)中积累框大小相同的重整框，将该重整窗中心从积累后图像中第一个像素开始不断移动，若重整框内中心像素的能量低于2a)中所确定的积累门限S_T，则对重整框内所覆盖的所有像素点做加1操作，否则，做加0操作；2b) Use the accumulation threshold S_T to uniformly reform the accumulated image sequence generated in 1c), that is, set a reformation frame with the same size as the accumulation frame in 1a), and change the center of the reforming window from the first one in the accumulated image. The pixels start to move continuously. If the energy of the center pixel in the reformation frame is lower than the accumulation threshold S_T determined in 2a), add 1 to all the pixels covered in the reformation frame, otherwise, add 0 operate;

2c)重复2b)中对重整窗中心的移动和累加操作直至重整窗中心点完全遍历图像序列，使得每个像素点完成N*N次累加运算，像素点的取值范围为[0,N*N]；2c) Repeat the movement and accumulation of the center of the reforming window in 2b) until the center of the reforming window completely traverses the image sequence, so that each pixel completes N*N accumulation operations, and the value range of the pixel is [0, N*N];

(3)对重整后的图像选取N*(N-1)/2为门限进行二值分割，分割后图像的取值范围为[0,1]；(3) Select N*(N-1)/2 as the threshold for binary segmentation of the reformed image, and the value range of the segmented image is [0,1];

(4)对(3)中二值分割后的图像进行连通域统计，删除图像中像素点个数大于2倍目标所占像素点总数的大型连通域和小于0.4倍目标所占像素点总数的小型连通域；(4) Perform connected domain statistics on the image after binary segmentation in (3), and delete the large connected domain in the image where the number of pixels is greater than 2 times the total number of pixels occupied by the target and those with less than 0.4 times the total number of pixels occupied by the target. small connected domains;

(5)对(4)处理后的图像进行帧间相关处理，完成运动目标的检测：(5) Inter-frame correlation processing is performed on the image processed in (4) to complete the detection of moving objects:

5a)初始化运动目标检测集，即将初始帧中的全部阴影检测结果作为初始运动目标阴影；5a) Initialize the moving target detection set, that is, take all the shadow detection results in the initial frame as the initial moving target shadow;

5b)逐帧对上一帧检测结果中相邻帧间位移超过运动目标尺寸一半的阴影进行保留，对相邻帧间位移不超过运动目标尺寸一半的阴影予以删除；5b) Retain the shadows whose displacement between adjacent frames exceeds half the size of the moving target in the detection result of the previous frame frame by frame, and delete the shadows whose displacement between adjacent frames does not exceed half the size of the moving target;

5c)对5b)保留的阴影中连续出现在3帧图像中的静止阴影和对在连续5帧图像中任意出现3帧的闪烁阴影均予以删除，最终得到动目标的检测结果。5c) Delete the static shadows that appear continuously in 3 frames of images and the flickering shadows that appear in 5 consecutive frames of images in 5b), and finally obtain the detection result of moving objects.

本发明与现有技术相比具有以下优点：Compared with the prior art, the present invention has the following advantages:

1.本发明通过对视频SAR图像序列进行逐像素点的扫描和检测，对每个像素点的联合区域信息进行判断，有效抑制了相干斑对检测的影响。1. In the present invention, the video SAR image sequence is scanned and detected pixel by pixel, and the joint area information of each pixel is judged, thereby effectively suppressing the influence of coherent speckles on detection.

2.本发明通过对积累图像进行二次阈值分割大幅度提高阴影区域的检测性能。2. The present invention greatly improves the detection performance of the shadow area by performing secondary threshold segmentation on the accumulated image.

3.本发明利用跟踪思想，避免了背景的提取，从而大大减少图像配准带来的计算，以及背景提取带来的边缘误差。3. The invention uses the tracking idea to avoid the extraction of the background, thereby greatly reducing the calculation caused by the image registration and the edge error caused by the background extraction.

4.本发明联系多帧间的动态信息，有效去除静态阴影及闪烁虚假运动目标，提高真实运动目标的检测准确度。4. The present invention links dynamic information between multiple frames, effectively removes static shadows and flickering false moving targets, and improves the detection accuracy of real moving targets.

附图说明Description of drawings

图1是本发明的实现流程图；Fig. 1 is the realization flow chart of the present invention;

图2是本发明中通过积累窗对图像进行区域积累的示意图；2 is a schematic diagram of performing regional accumulation on an image by an accumulation window in the present invention;

图3是本发明中对图像进行统一重整的示意图；Fig. 3 is the schematic diagram of carrying out unified reformation to the image in the present invention;

图4是用本发明进行视频SAR运动目标检测的仿真结果图。FIG. 4 is a simulation result diagram of video SAR moving target detection using the present invention.

具体实施方式Detailed ways

下面结合附图对本发明的实施例和效果做进一步描述。The embodiments and effects of the present invention will be further described below with reference to the accompanying drawings.

参照图1，本发明的实施步骤如下：1, the implementation steps of the present invention are as follows:

步骤一，设计积累窗并对高帧率视频SAR图像序列进行区域积累。Step 1: Design an accumulation window and perform regional accumulation of high frame rate video SAR image sequences.

1.1)设置积累窗大小N：1.1) Set the accumulation window size N:

积累窗的大小由视频SAR图像序列的分辨率以及运动目标在图像中的尺寸决定，过小的窗尺寸不能有效对阴影区域进行积累，过大的窗尺寸将恶化图像序列的分辨率。积累窗的设计规则是要与运动目标阴影尺寸大小相当且不大于运动目标阴影尺寸，以保证存在部分阴影区域能被积累窗完全覆盖，且考虑到运动目标的运动方向无法先验获取，故采用正方形积累窗口。The size of the accumulation window is determined by the resolution of the video SAR image sequence and the size of the moving object in the image. A window size that is too small cannot effectively accumulate shadow areas, and a window size that is too large will deteriorate the resolution of the image sequence. The design rule of the accumulation window is to be equal to the shadow size of the moving target and not larger than the shadow size of the moving target, so as to ensure that some shadow areas can be completely covered by the accumulation window, and considering that the moving direction of the moving target cannot be obtained a priori, it is adopted. Square accumulation window.

本实例中，是在视频SAR图像序列中选取第104帧如图4(a)所示，SAR图像分辨率为0.2m，运动目标尺寸为1.8m*4.6m，故运动目标阴影将至少占据SAR图像中9*23个像素点，为保证阴影区域得到有效积累，取N＝2*ceil(min(N₁,N₂)/2)-1，其中ceil()表示向上取整，*表示点乘，N₁,N₂为运动目标在图像中占据的二维像素点大小；In this example, the 104th frame is selected from the video SAR image sequence, as shown in Figure 4(a). The resolution of the SAR image is 0.2m, and the size of the moving target is 1.8m*4.6m, so the shadow of the moving target will occupy at least SAR. There are 9*23 pixels in the image, in order to ensure the effective accumulation of the shadow area, take N=2*ceil(min(N₁ , N₂ )/2)-1, where ceil() means rounding up, * means point Multiply, N₁ , N₂ is the two-dimensional pixel size occupied by the moving target in the image;

在本例中N₁＝9,N₂＝23，则取N＝7，则积累窗的尺寸设计为7*7；In this example, N₁ =9, N₂ =23, then taking N=7, the size of the accumulation window is designed to be 7*7;

1.2)使用滑窗方法进行图像区域积累，得到当前积累窗中心的强度：1.2) Use the sliding window method to accumulate the image area to obtain the intensity of the current accumulation window center:

如图2所示，本步骤是利用积累窗通过逐一遍历像素点的方法对视频SAR新图像进行区域积累，即先将积累窗中心从第一个图像像素开始不断移动，使得积累窗在移动过程中覆盖不同的图像区域，导致积累窗内每个像素点的强度大小随之改变；再将积累窗所覆盖区域所有像素点的强度进行求和，得到当前积累窗中心的强度。As shown in Figure 2, this step is to use the accumulation window to traverse the pixel points one by one to perform regional accumulation on the new video SAR image, that is, the center of the accumulation window is continuously moved from the first image pixel, so that the accumulation window is in the moving process. The intensity of each pixel in the accumulation window changes accordingly; then the intensities of all pixels in the area covered by the accumulation window are summed to obtain the intensity at the center of the current accumulation window.

1.3)重复1.2)操作直至积累窗中心完全遍历视频SAR图像序列，得到积累后的图像序列。1.3) Repeat the operation of 1.2) until the center of the accumulation window completely traverses the video SAR image sequence to obtain the accumulated image sequence.

积累后的图像如图4(b)所示。The accumulated image is shown in Fig. 4(b).

步骤二，确定积累门限S_T。Step 2, determine the accumulation threshold S_T .

积累窗内的积累强度由SAR图像中背景强度、噪声水平以及阴影区域强度综合决定，给定虚警率和检测率要求，可导出区域积累后图像强度门限S_T。The accumulation intensity in the accumulation window is comprehensively determined by the background intensity, noise level and shadow area intensity in the SAR image. Given the requirements of false alarm rate and detection rate, the image intensity threshold S_T after area accumulation can be derived.

2.1)对于均匀背景区域，每个像素点I_i的图像强度值服从独立同分布的指数分布I_i～exp(λ_i)，其概率密度函数为：2.1) For a uniform background area, the image intensity value of each pixel point I_i obeys the independent and identically distributed exponential distribution I_i ~exp(λ_i ), and its probability density function is:

其中λ_i＝1/σ_i表示第i个地面分辨单元的总体等效后向散射系数σ_i的倒数；where λ_i =1/σ_i represents the reciprocal of the overall equivalent backscattering coefficient σ_i of the ith ground resolving unit;

2.2)选取积累窗内的7*7个分辨单元，则积累窗内每个像素点的积累强度S表示为:2.2) Select 7*7 resolution units in the accumulation window, then the accumulation intensity S of each pixel in the accumulation window is expressed as:

2.3)将积累窗内的49个分辨单元强度S分成n组，每组具有相同的参数λ_i，计算积累强度的概率密度函数为：2.3) Divide the 49 resolution unit intensities S in the accumulation window into n groups, each group has the same parameter λ_i , and the probability density function for calculating the accumulated intensities is:

其中λ_i,p_i分别表示指数分布的参数以及对应该参数的随机变量的数量，m_l为符合m₁+...+m_n＝p_i-j和m_i≠0条件的所有自然数，l＝1,2...n；where λ_i , p_i represent the parameters of exponential distribution and the number of random variables corresponding to the parameters, respectively, m_l is all natural numbers that meet the conditions of m₁ +...+m_n =p_i -j and m_i ≠0, l=1,2...n;

2.4)计算在式<4>概率密度函数条件下的检测概率P_d为：2.4) Calculate the detection probability P_d under the condition of the probability density function of formula <4>:

2.5)根据统计学原理，计算背景为均匀区域时服从Gamma分布的积累强度的概率密度函数：2.5) According to statistical principles, calculate the probability density function of the cumulative intensity of the Gamma distribution when the background is a uniform area:

其中Γ(N)表示参数为积累窗大小N的伽马函数；where Γ(N) represents the gamma function whose parameter is the accumulation window size N;

2.6)给定积累门限S_T，计算阴影区域的虚警概率：2.6) Given the accumulation threshold S_T , calculate the false alarm probability of the shadow area:

2.7)利用式<5>、式<7>和背景散射强度参数λ_i，在不低于90％的检测概率P_d及不高于1％的虚警率P_fa的条件下，选取积累门限S_T。2.7) Using formula <5>, formula <7> and the background scattering intensity parameter λ_i , select the accumulation threshold under the condition that the detection probability P_d is not lower than 90% and the false alarm rate P_fa is not higher than 1%_ST .

在本例中，选取积累门限为5。In this example, the accumulation threshold is chosen to be 5.

步骤三，进行区域图像重整。The third step is to perform regional image reorganization.

3.1)设置与1.1)中积累框大小相同的重整框，将该重整窗中心从积累后图像中的第一个像素开始不断移动，在每次移动后，若重整框中心像素的图像总能量低于步骤二中所确定的积累门限S_T，则对重整框内所覆盖的所有像素点做加1操作，否则，做加0操作，如图3所示；3.1) Set a reformation frame with the same size as the accumulation frame in 1.1), and move the center of the reformation window continuously from the first pixel in the accumulated image. If the total energy is lower than the accumulation threshold S_T determined in step 2, add 1 to all the pixels covered in the reformation frame, otherwise, add 0, as shown in Figure 3;

3.2)重复3.1)操作，直至重整窗中心完全遍历积累图像序列，完成整幅图像的重整后，每个像素将完成7*7次累加运算，其取值范围为[0,49]，重整后的图像如图4(c)所示。3.2) Repeat 3.1) until the center of the reforming window completely traverses the accumulated image sequence. After the entire image is reformed, each pixel will complete 7*7 accumulation operations, and its value range is [0,49], The reformed image is shown in Fig. 4(c).

步骤四，对步骤三中重整后的图像进行二值分割。In step 4, binary segmentation is performed on the image reformed in step 3.

对重整后的图像选取N*(N-1)/2为门限，对像素值大于等于该门限的像素单元进行赋1操作，对像素值小于该门限的像素单元进行赋0操作，完成二值分割。Select N*(N-1)/2 as the threshold for the reformed image, assign 1 to the pixel units whose pixel value is greater than or equal to the threshold, and assign 0 to the pixel units whose pixel value is less than the threshold. value split.

本实例中设定分割门限为21，分割后结果如图4(d)所示。从图4(d)可见，分割后的图像存在成片的低强度区域以及一些因静止目标高度产生的固定阴影。为减少后续步骤的运算量，应设法减少此类由非运动目标产生的阴影的数量。In this example, the segmentation threshold is set to 21, and the result after segmentation is shown in Figure 4(d). It can be seen from Figure 4(d) that the segmented image has patches of low-intensity areas and some fixed shadows due to the height of the stationary object. In order to reduce the computational complexity of the subsequent steps, we should try to reduce the number of such shadows produced by non-moving objects.

步骤五，对二值分割后的图片进行连通域阴影统计并进行处理。Step 5: Perform connected domain shadow statistics on the image after binary segmentation and process it.

连通域分为单连通域和多连通域，其单连通域的定义为：对于复平面上的一块区域G，在区域中任做一条简单闭曲线，如果闭曲线的内部全部属于区域G，则称G为单连通域，任何不是单连通域的区域都是多联通域。Connected domains are divided into single-connected domains and multi-connected domains. The definition of a single-connected domain is: for a region G on the complex plane, make any simple closed curve in the region. If the interior of the closed curve belongs to the region G, then G is called a simply connected domain, and any region that is not a single connected domain is a multi-connected domain.

上述步骤四得到分割后图像内的连通域多为单连通域，不会出现多连通域，如本例的图4(d)所示，但在某些极特殊情况下，分割后的图像内会出现多连通域，由于多连通域都不由动目标产生，能在之后的帧间相关处理中得到清除，所以在此步骤中只关心连通域大小，不对单连通域和多连通域进行区分。The connected domain in the segmented image obtained in the above step 4 is mostly a single connected domain, and there will be no multi-connected domain, as shown in Figure 4(d) in this example, but in some very special cases, the segmented image There will be multi-connected domains. Since the multi-connected domains are not generated by the moving target, they can be cleared in the subsequent inter-frame correlation processing. Therefore, in this step, only the size of the connected domain is concerned, and the single-connected domain and the multi-connected domain are not distinguished.

本步骤的具体实现是：首先找出所有连通域，把每一块连通域在图像中占据的像素点总数作为当前连通域的大小，再根据目标在图像中所占像素点总数N₁*N₂，保留大小在[0.4*N₁*N₂,2*N₁*N₂]范围内的连通域。The specific implementation of this step is: firstly find all connected domains, take the total number of pixels occupied by each connected domain in the image as the size of the current connected domain, and then according to the total number of pixels N₁ *N₂ occupied by the target in the image , preserves the connected domain whose size is in the range of [0.4*N₁ *N₂ ,2*N₁ *N₂ ].

在本实例中目标在图像上所占分辨单元为200个，故在这里选筛选出像素数量在[80,400]之间的连通域作为候选运动目标阴影区域，筛选后的检测结果如图4(e)所示。In this example, the number of resolution units occupied by the target on the image is 200, so the connected domain with the number of pixels between [80, 400] is selected as the candidate moving target shadow area, and the filtered detection result is shown in Figure 4 (e ) shown.

步骤六，对连通域处理后的图像进行帧间相关处理去除非目标阴影，完成运动目标检测。Step 6: Perform inter-frame correlation processing on the image processed in the connected domain to remove non-target shadows and complete moving target detection.

在完成对连通域阴影大小的筛选后，还会留下部分与运动目标阴影大小相近、形状相似的固定阴影，此类阴影无法从单帧图像中判定是否为运动目标阴影，因此需要通过多帧图像间的关系区分静止阴影和运动目标阴影，并对静止阴影进行去除。此外在筛选的过程中，部分阴影的像素数量处于[80,400]的边缘处，导致在某些图像帧中此区域能够被确定为运动目标阴影，与运动目标的阴影不同，此类固定阴影在观测序列中的相同位置间断出现，在连续帧中产生阴影闪烁的现象，也应对此类闪烁阴影予以消除。After the screening of the shadow size of the connected domain is completed, some fixed shadows with similar size and shape to the shadow of the moving target will be left. Such shadows cannot be determined from a single frame of images as shadows of moving targets, so it is necessary to pass multiple frames of shadows. The relationship between images distinguishes static shadows from moving target shadows, and removes static shadows. In addition, in the process of screening, the number of pixels of partial shadows is at the edge of [80, 400], resulting in this area can be determined as shadows of moving objects in some image frames. Different from shadows of moving objects, such fixed shadows are observed Intermittent occurrences of the same position in the sequence, resulting in shadow flickering in successive frames, should also be eliminated.

对非目标阴影的排除的已有方法中通常有背景差分法，光流法和航迹关联法。本实例采用但不限于航迹关联方法，具体实施步骤如下：The existing methods for excluding non-target shadows usually include background difference method, optical flow method and track correlation method. This example adopts but is not limited to the track association method, and the specific implementation steps are as follows:

6.1)初始化运动目标检测集，即将初始帧中的全部阴影检测结果作为初始运动目标阴影；6.1) Initialize the moving target detection set, that is, take all the shadow detection results in the initial frame as the initial moving target shadow;

6.2)逐帧对上一帧检测结果中相邻帧间位移不超过目标尺寸一半的阴影进行保留，对相邻帧间位移超过目标尺寸一半的阴影予以删除；6.2) Retain the shadows whose displacement between adjacent frames does not exceed half of the target size in the detection result of the previous frame frame by frame, and delete the shadows whose displacement between adjacent frames exceeds half the target size;

6.3)对6.2)中保留的阴影中连续出现在3帧图像中的静止阴影和对在连续5帧图像中任意出现3帧的闪烁阴影均予以删除，最终得到动目标的检测结果，如图4(g)中白色方框所示。6.3) Delete the static shadows that appear continuously in 3 frames of images in the shadows retained in 6.2) and the flickering shadows that appear in any 3 frames in 5 consecutive frames of images. Finally, the detection results of moving objects are obtained, as shown in Figure 4 In (g), the white box is shown.

以上描述仅是本发明的一个具体事例，并未构成对本发明的任何限制，显然对于本领域的专业人员来说，在了解本发明内容和原理后，都可能在不背离本发明原理、结构的情况下，进行形式和细节上的各种修改和改变，但是这些基于本发明思想的修正和改变仍在本发明的权利要求保护范围之内。The above description is only a specific example of the present invention, and does not constitute any limitation to the present invention. Obviously, for those skilled in the art, after understanding the content and principles of the present invention, they may not deviate from the principles and structures of the present invention. Under certain circumstances, various modifications and changes in form and details are made, but these modifications and changes based on the idea of the present invention are still within the protection scope of the claims of the present invention.

Claims (5)

1. A video SAR moving target detection method based on image region accumulation is characterized by comprising the following steps:

(1) carrying out regional accumulation on a video SAR image sequence:

1a) setting an accumulation window, and determining the size of the accumulation window to be N x N according to the resolution of the video SAR image sequence and the size of the shadow of the moving target in the image, wherein N is an odd number and x represents dot multiplication;

1b) accumulating the image area by using a sliding accumulation window method, wherein the center of the accumulation window continuously moves from a first image pixel, and the value of each corresponding pixel point is determined by summing the image sequence values covered in the accumulation window around the point;

1c) repeating the step 1b) until the center of the accumulation window completely traverses the video SAR image sequence to generate an accumulated image sequence;

(2) determining an accumulation threshold S_TAnd performing image reformation on the accumulated result:

2a) according to a uniform background probability density function f_B(s), system noise level λ and false alarm rate P_faAnd a detection probability P_dIn combination with the following equation, determining an accumulation threshold S_TNamely:

2b) using an accumulation threshold S_TUniformly reforming the accumulated image sequence generated in 1c), namely setting a reforming frame with the same size as the accumulation frame in 1a), continuously moving the center of the reforming window from the first pixel in the accumulated image, and if the pixel energy at the center of the reforming frame is lower than the accumulation threshold S determined in 2a)_TIf not, adding 0 to all the pixel points covered in the reforming frame;

2c) repeating the moving and accumulation operations of the center of the reforming window in the step 2b) until the center of the reforming window completely traverses the image sequence, so that each pixel point completes N times of accumulation operation, and the value range of the pixel point is [0, N times N ];

(3) selecting N x (N-1)/2 as a threshold for binary segmentation of the reformed image, wherein the value range of the segmented image is [0,1 ];

(4) performing connected domain statistics on the image subjected to binary segmentation in the step (3), and deleting large connected domains with the number of pixel points being more than 2 times of the total number of the pixel points occupied by the target and small connected domains with the number of the pixel points being less than 0.4 time of the total number of the pixel points occupied by the target in the image;

(5) and (5) performing inter-frame correlation processing on the image processed in the step (4) to finish detection of a moving target:

5a) initializing a moving target detection set, namely taking all shadow detection results in an initial frame as initial moving target shadows;

5b) retaining the shadow of which the displacement between adjacent frames in the detection result of the previous frame does not exceed half of the target size frame by frame, and deleting the shadow of which the displacement between adjacent frames exceeds half of the target size;

5c) and deleting the static shadows continuously appearing in the 3 frames of images in the shadows reserved in the step 5b) and the flicker shadows randomly appearing in the 3 frames of images in the continuous 5 frames of images, and finally obtaining the detection result of the moving target.

2. The method according to claim 1, wherein the determination of the size of the accumulation window in step 1a) as a function of the resolution of the sequence of video SAR images and of the size of the shadow of the moving object in the images is based on the resolution determined by the radar system and the N occupied by the moving object in the images₁*N₂A resolution unit for determining the size of the accumulation window as N × N, wherein N is 2 × ceil (min (N)₁,N₂) /2) -1, ceil () represents rounding up.

3. The method according to claim 1, wherein the step 1b) of accumulating the image areas by using a sliding accumulation window method comprises the steps of moving the center of the accumulation window from the first image pixel continuously, so that the accumulation window covers different image areas in the moving process, the intensity of each pixel point in the accumulation window is changed accordingly, and then summing the intensities of all the pixel points in the area covered by the accumulation window to obtain the intensity of the center of the current accumulation window.

4. The method according to claim 1, wherein the accumulation threshold is determined in step 2a) as follows:

2a1) calculating the intensity I of each resolution unit of the accumulation window according to the background intensity, the noise level and the shadow region intensity in the SAR image_iProbability density function of (1):

wherein λ_i＝1/σ_iRepresenting the overall equivalent backscattering coefficient σ of the ith ground resolution element_iThe reciprocal of (a);

2a2) selecting N × N resolution cells within the accumulation window, the accumulation intensity can be expressed as:

2a3) will be provided withThe N2 resolution cell intensities within the accumulation window are divided into N groups, each group having the same parameter λ_iThe probability density function for calculating the accumulation intensity is:

wherein λ_i,p_iNumber of parameters, m, respectively representing the exponential distribution and the random variables of the parameters_lAll conform to m₁+...+m_n＝p_i-j and m_iA natural number not equal to 0 condition, l being 1,2.. n;

2a4) the target detection probability under the condition of the probability density function of the formula <4> is:

2a5) according to the statistical principle, the accumulated intensity obeying the Gamma distribution when the background is a uniform area is calculated, and the probability density function is:

wherein Γ (N) is a gamma function with a parameter N;

2a6) given accumulation threshold S_TThe false alarm probability of the shadow region is:

by using the above formula<5>、<7>And a background scattering intensity parameter λ_iAnd defining a detection probability P not less than 90%_dAnd a false alarm rate P of not more than 1%_faUnder the condition of (1), selecting an accumulation threshold S_T。

5. The method according to claim 1, wherein the step (4) of performing connected component statistics on the binary-segmented picture is to perform processing on each binary-segmented pictureThe total number of pixel points occupied by one connected domain in the image is used as the size of the current connected domain, and the total number N of the pixel points occupied by the target in the image is used as the size of the current connected domain₁*N₂With a size of the connected component of [ 0.4N ]₁*N₂,2*N₁*N₂]Of the connected domain.

Images