CN105988113A - Polarmetric synthetic aperture radar (SAR) image change detection method - Google Patents

Abstract

The invention belongs to the technical field of SAR image processing. Through in-depth interpretation of polarimetric SAR data, a method for quickly and accurately extracting two-time phase change regions is studied, and a method for detecting changes in polarimetric SAR images based on joint weighted polarization differences is realized. . The technical scheme adopted in the present invention is a method for detecting changes in polarization synthetic aperture radar images, the steps of which are as follows: (1) preprocessing, performing de-orientation and coherent speckle noise suppression operations on the registered two-temporal polarization SAR images; Two) Construct the feature vectors k _Ai and k _Bi of the corresponding pixels in the two-time phase image; (3) Calculate the polarization scattering difference and the polarization power difference of the corresponding pixel in the two-time phase image; (4) According to the two The relative size of the difference degree is assigned the corresponding weighting coefficient, and the joint weighted polarization difference degree is obtained after summing up, and the difference image is constructed; (5) The difference image is thresholded to extract the change area. The invention is mainly applied to image processing occasions.

Description

A Method for Detection of Changes in Polarimetric Synthetic Aperture Radar Images

technical field

The invention belongs to the technical field of SAR image processing, and relates to a polarization SAR image change detection method based on a joint weighted polarization difference degree.

Background technique

Synthetic Aperture Radar (SAR) is an all-day, all-weather high-resolution microwave imaging system. Because it uses microwave technology for imaging, it can effectively get rid of the limitations of natural factors such as weather and achieve real-time observation. These characteristics It also has unique advantages in natural disaster monitoring, ocean observation, and military reconnaissance. The traditional single-channel single-polarization SAR image can only obtain the target scattering characteristics of the ground scene under a specific polarization transceiver combination, and the amount of information is very limited. In recent years, the SAR system that can obtain the polarization scattering characteristics of ground objects (that is, the polarimetric SAR system) has been continuously developed, and the imaging technology of polarimetric synthetic aperture radar is also becoming more and more mature. Compared with single-polarization SAR images, multi-polarization/full-polarization SAR images contain more information, and can more completely and accurately reveal the scattering mechanism of the target. In order to fully explore the target information in the image, improve segmentation, classification, target The detection and identification performance provides data support and guarantee. Due to the complexity of the imaging mechanism of the polarimetric SAR system and its late start, it is not easy to interpret the polarimetric SAR images reasonably and correctly, which also makes the research in the related fields of polarimetric SAR image processing have a lot of room for development. As an important branch of SAR image interpretation, change detection technology based on Polarimetric Synthetic Aperture Radar (PolSAR) images plays a pivotal role in many fields such as disaster estimation, urban planning, and military strike.

In recent years, scholars at home and abroad have done quite in-depth research on the change detection technology of SAR images, and proposed many methods, such as ratio method, difference method, classification comparison method, vegetation index method and so on. Although scholars at home and abroad have made some progress in multi-polarization SAR image change detection, the research is not perfect and mature, and a unified system has not been formed. In 2003, Knut Conradsen et al. from the International Space Research Institute of the Technical University of Denmark proposed a change detection method for polarimetric SAR images based on the complex Wishart distribution. In 2004, Japanese scholar Muhtar Qong proposed a method for detecting changes in polarization SAR images based on polarization state construction. This method improves the accuracy of change detection by constructing an optimal polarization state. In 2008, Laurent Ferro-Famil realized the change detection by using the maximum likelihood ratio (Maximum Likelihood ratio, MLR) to construct the change detection feature quantity. In 2011, Esra Erten et al. used the KL distance as the change detection feature to realize the change detection of polarimetric SAR images. In 2013, Bhogendra Mishra et al. proposed a new feature quantity for change detection——Normalized Difference Ratio (NDR). The ratio detection amount is more, and the detection effect is better.

Contents of the invention

In order to overcome the deficiencies of the prior art, the present invention aims to study a fast and accurate method for extracting two-temporal phase change regions through in-depth interpretation of the polarimetric SAR data, and realize the polarimetric SAR image based on the joint weighted polarization difference degree Change detection method. The technical scheme adopted in the present invention is a method for detecting changes in polarization synthetic aperture radar images, the steps of which are as follows:

(1) Preprocessing, performing de-orientation and coherent speckle noise suppression operations on the registered two-temporal polarization SAR images;

(2) Construct the feature vectors k _Ai and k _Bi of the corresponding pixels in the two-temporal image

k _Ai = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

k _Bi = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

Among them, k _Ai and k _Bi represent the eigenvectors of the pixel point i in the two-time polarimetric SAR data of A and B respectively, T _mn represents the elements in the coherence matrix T of the polarimetric SAR image, and the superscript T means transpose;

(3) Calculate the polarization scattering difference degree and polarization power difference degree of the corresponding pixel points of the two-temporal images

Assuming that the characteristic vectors of the two targets are k _Ai and k _Bi respectively, and the corresponding antenna receiving powers are P _Ai and P _Bi respectively, the polarization scattering difference and polarization power difference between the two targets are defined as:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}$

Among them, ||·|| ₂ represents the 2-norm of the vector, and the superscript H represents the conjugate transpose;

(4) Assign corresponding weighting coefficients according to the relative size of the two differences, and obtain the joint weighted polarization difference after summing, and construct a difference image. The joint weighted polarization difference is:

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>$

Among them, a and b are weighting coefficients and satisfy a+b=1, a and b determine the contribution of the two difference types to the overall difference;

(5) Perform threshold segmentation on the difference image to extract the change area: first, use the change detection feature quantity, that is, the mean and standard deviation of the difference image to determine the candidate threshold, and then determine the final segmentation threshold according to the ratio of the number of change points to the number of all pixels , and finally use the threshold segmentation method to extract the changed region.

Features and beneficial effects of the present invention are:

The present invention comprehensively utilizes the polarization information of the polarimetric SAR image, takes the joint weighted polarization difference degree as the change detection measure, and provides a polarization SAR image change detection method based on the joint weighted polarization difference degree, which can be Fast and effective detection of changing regions in two-temporal images.

Description of drawings:

Figure 1 shows the two-temporal Pauli decomposition diagram of farmland data.

Figure 2 shows the difference image of the method of the present invention.

Fig. 3 shows the result of threshold segmentation by the method of the present invention.

Fig. 4 shows a flowchart of the method proposed by the present invention.

detailed description

The present invention makes full use of the information of the polarimetric SAR image, and realizes a method for detecting the change of the polarimetric SAR image based on the joint weighted polarization difference degree. The specific technical scheme is divided into the following steps:

(1) Pretreatment. It mainly includes de-orientation and coherent speckle noise suppression operations on the registered two-temporal polarimetric SAR images, so as to reduce the impact of random orientation of ground objects and coherent speckle noise on the change detection results.

(2) Construct feature vectors k _Ai and k _Bi of corresponding pixels in the two-temporal image.

k _Ai = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

k _Bi = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

Among them, k _Ai and k _Bi represent the eigenvectors of pixel i in A and B polarimetric SAR data respectively, T _mn represents the elements in the coherence matrix T of the polarimetric SAR image, and the superscript T represents transposition.

(3) Calculate the polarization scattering difference degree and polarization power difference degree of the corresponding pixel points of the two-temporal images.

Assuming that the characteristic vectors of the two targets are k _Ai and k _Bi respectively, and the corresponding antenna receiving powers are P _Ai and P _Bi respectively, the polarization scattering difference and polarization power difference between the two targets are defined as:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}$

where ||·|| ₂ denotes the 2-norm of the vector, and the superscript H denotes the conjugate transpose.

(4) Assign corresponding weighting coefficients according to the relative size of the two difference degrees, and obtain the joint weighted polarization difference degree after summing, and construct the difference image. The joint weighted polarization difference is:

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>$

Among them, a and b are weighting coefficients and satisfy a+b=1, a and b determine the contribution of the two difference types to the overall difference degree.

(5) Perform threshold segmentation on the difference image to extract the change area. Firstly, the mean and standard deviation of the change detection feature (i.e., the difference image) are used to determine the candidate threshold, and then the final segmentation threshold is determined according to the ratio of the number of change points to the number of all pixels, and finally the threshold segmentation method is used to extract the change area.

The present invention provides a new weighting coefficient selection scheme:

Theoretically speaking, under the condition of constant imaging conditions, the scattering characteristics and power characteristics of the same surface object in different time phase images should be the same. When the ground object at the same position in the two-phase images changes, its scattering characteristics and power characteristics should also change correspondingly, but the degree of change is not necessarily the same, and the scattering characteristics may change significantly, and the power characteristics may also change significantly. Therefore, when assigning weighting coefficients a and b, a compromise should be considered according to the relative size of the two differences. Based on the above considerations, the selection scheme of the weighting coefficient is as follows:

When D ₁ >D ₂ , the principle of the weighting coefficient assigned is: a<0.5, b>0.5;

When D ₁ ≤ D ₂ , the principles of the assigned weighting coefficients are: a>0.5, b<0.5.

The present invention proposes an automatic threshold value selection method that integrates data histogram distribution and change point ratio. The main idea is to determine the candidate threshold value through the mean and standard deviation of the difference image, and then iteratively determine the final threshold value according to prior information such as the change point ratio. Segmentation threshold. Specific steps are as follows:

(1) Initialize parameters. Let the ratio of preset change points be M, and the number of iterations be

(2) Calculate the mean μ and standard deviation σ of the entire difference image;

(3) Determine that the threshold in this iterative cycle is

(4) Perform threshold segmentation on the difference image to determine the actual change ratio M';

(5) If the iteration termination condition M′≤M is not satisfied, update And repeat steps (3-5) until M'≤M is satisfied, at this time, the segmentation threshold Th is output.

The present invention uses the joint weighted polarization difference degree to describe the difference degree of the corresponding target in the two-temporal images, and the measure can realize the emphatic selection of the difference type by adjusting the size of the weighting coefficient. First calculate the polarization scattering difference and polarization power difference of the corresponding targets at two times, then determine the weighting coefficient according to the weighting coefficient selection scheme proposed by the present invention, and obtain the joint weighting between the corresponding targets after weighting and summing the two parts of the difference The degree of polarization difference, and finally the threshold segmentation technology is used to extract the change area to realize the change detection. The implementation process of the polarization SAR image change detection method based on the joint weighted polarization difference proposed by the present invention is introduced below.

1. Pretreatment.

This step mainly includes de-orientation and coherent speckle noise suppression operations on the registered two-temporal polarimetric SAR images, so as to reduce the impact of random orientation of ground objects and coherent speckle noise on the change detection results. As we all know, speckle noise in polarimetric SAR images is unavoidable. Therefore, speckle suppression, that is, denoising in the traditional sense, is required before change detection to reduce the impact of speckle noise on experimental results. The present invention utilizes the speckle reduction algorithm based on the scattering model proposed by Lee in 2006 to suppress coherent speckle on the image. De-orientation refers to removing the influence of target orientation on scattering to highlight the essential characteristics of the target. Two scattering targets with different orientations and other features are exactly the same. After this operation, the polarization scattering information should be completely consistent. Taking the de-orientation operation as a process in the pre-processing of change detection can effectively avoid the influence of random orientation on the experimental results, and is conducive to enhancing the reliability and accuracy of the experimental results.

2. Constructing feature vectors k _Ai and k _Bi of corresponding pixels in the two-temporal image.

The variable polarization effect of the target is usually characterized by the polarization coherence matrix T that contains all the polarization information of the target, and its expression is as follows:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{ccc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 13</span>}}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; three</span>}}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 32</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{ccc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 13</span>}}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}^{<span\; class="notranslate">\; *</span>}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; three</span>}}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 13</span>}}^{<span\; class="notranslate">\; *</span>}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; three</span>}}^{<span\; class="notranslate">\; *</span>}& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}\end{array}\right]$

Wherein, T _mn represents an element in the polarization coherence matrix T, and the superscript * represents a conjugate operation.

Vectorize the T matrix to get:

K＝[T ₁₁ T ₁₂ T ₁₃ T ₂₁ T ₂₂ T ₂₃ T ₃₁ T ₃₂ T ₃₃ ] ^T

Among them, the superscript T means transpose. Considering that the T matrix is a Hermite matrix, that is, the upper triangular elements and lower triangular elements of the matrix satisfy the relationship of mutual conjugation, therefore, only the upper triangular elements of the T matrix can be used to construct the eigenvector, and its expression is as follows:

k＝[T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

Obviously, the vector K and the vector k contain the same target information, but the k vector is a six-dimensional vector, and the calculation amount can be reduced and the calculation efficiency can be improved when the vector is used for calculation. Corresponding to the two-temporal polarization SAR image, we express it as the following formula:

k _Ai = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

k _Bi = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T

Among them, k _Ai and k _Bi represent the feature vectors of the pixel point i in the polarimetric SAR data of the two time phases A and B, respectively.

3. Calculate the degree of scattering difference and the degree of power difference.

Suppose the eigenvectors of the corresponding pixels of the two-phase polarimetric SAR image are k _Ai and k _Bi respectively, and the corresponding antenna receiving powers are P _Ai and P _Bi respectively, define the polarization scattering difference and polarization power difference between two targets degrees are:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}$

Among them, ||·|| ₂ represents the 2-norm of the vector, and the superscript H represents the conjugate transpose operation. The polarization scattering difference describes the correlation of two coherence matrices, and the polarization power difference describes the echo power difference of two targets.

4. Calculate the joint weighted polarization difference degree.

The corresponding weighting coefficients are allocated according to the relative size of the two difference degrees, and the joint weighted polarization difference degree is obtained after the summation, and the difference image is constructed. The joint weighted polarization difference is:

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>$

Among them, a and b are weighting coefficients and satisfy a+b=1, a and b determine the contribution of the two difference types to the overall difference degree. The degree of polarization difference describes the degree of difference between two targets. The greater the value of the degree of difference, the greater the difference between the two targets; on the contrary, the smaller the value of the degree of difference, the smaller the degree of difference between the two targets.

Statistically comparing the numerical characteristics of the scattering difference and power difference of multiple sets of data, it is found that not all pixels have a uniform absolute dominant difference type, so giving all pixels the same weighting coefficient will lose part of the suboptimal difference type information . In view of this, the present invention proposes a new weighting coefficient distribution scheme. Theoretically speaking, under the condition of constant imaging conditions, the scattering characteristics and power characteristics of the same surface object in different time phase images should be the same. When the ground object at the same position in the two-phase images changes, its scattering characteristics and power characteristics should also change correspondingly, but the degree of change is not necessarily the same, and the scattering characteristics may change significantly, and the power characteristics may also change significantly. Therefore, when assigning weighting coefficients a and b, a compromise should be considered according to the relative size of the two differences. Based on the above considerations, the weighting coefficient selection scheme we determined is as follows:

When D ₁ >D ₂ , the principle of the weighting coefficient assigned is: a<0.5, b>0.5;

When D ₁ ≤ D ₂ , the principles of the assigned weighting coefficients are: a>0.5, b<0.5.

5. Threshold segmentation to extract changing regions.

In this step, the segmentation threshold is firstly determined by using the automatic threshold selection method of comprehensive data histogram distribution and change point ratio proposed by the present invention, and then using the segmentation threshold to segment the difference image to obtain the final change detection result.

The present invention proposes an automatic threshold selection method that integrates data histogram distribution and change point ratio. The main idea is to determine the candidate threshold through the mean and standard deviation of the difference image, and then iteratively determine the final segmentation according to prior information such as the change point ratio. threshold. This threshold selection method is inspired by the concept of the confidence interval of the Gaussian distribution. The mean value of the difference image represents the central distribution of the entire difference to be segmented. There are a large number of false alarms, so the standard deviation needs to be accumulated on the basis of the mean value. As for the standard deviation of accumulating several times, it can be determined according to the ratio of the change point. Specific steps are as follows:

(1) Initialize parameters. Let the ratio of preset change points be M, and the number of iterations be

(2) Calculate the mean μ and standard deviation σ of the entire difference image;

(3) Determine that the threshold in this iterative cycle is

(4) Perform threshold segmentation on the difference image to determine the actual change ratio M';

(5) If the iteration termination condition M′≤M is not satisfied, update And repeat steps (3-5) until M'≤M is satisfied, at this time, the segmentation threshold Th is output.

The experimental results are illustrated below in conjunction with the accompanying drawings:

Figure 1 shows the Pauli decomposition diagram of two-phase polarimetric SAR data collected over Kings County, California, USA. The left image was collected on May 19, 2011, and the right image was collected on May 20, 2011. Since mid-May is the busy season in the local area, although the data are only one day apart, there are obvious changes caused by crop irrigation, sowing and tillage. Seven obvious change areas are marked in the figure. Figure 2 and Figure 3 show the difference images and the final change detection results of the method of the present invention. It can be seen from the figures that the present invention can effectively extract the change area with less false alarms and clear outlines.

Claims (1)

1. A polarization synthetic aperture radar image change detection method is characterized in that the steps are as follows: (1) Preprocessing, performing de-orientation and coherent speckle noise suppression operations on the registered two-temporal polarization SAR image; (2) Construct the feature vectors k _Ai and k _Bi of the corresponding pixels in the two-temporal image k _Ai = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T k _Bi = [T ₁₁ T ₁₂ T ₁₃ T ₂₂ T ₂₃ T ₃₃ ] ^T Among them, k _Ai and k _Bi represent the eigenvectors of pixel i in A and B polarimetric SAR data respectively, T _mn represents the elements in the coherence matrix T of the polarimetric SAR image, and the superscript T represents the transformation set; (3) Calculate the polarization scattering difference degree and polarization power difference degree of the corresponding pixel points of the two-temporal images Assuming that the characteristic vectors of the two targets are k _Ai and k _Bi respectively, and the corresponding antenna receiving powers are P _Ai and P _Bi respectively, the polarization scattering difference and polarization power difference between the two targets are defined as: ${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}$ ${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}$ Among them, ||·|| ₂ represents the 2-norm of the vector, and the superscript H represents the conjugate transpose; (4) Assign corresponding weighting coefficients according to the relative size of the two differences, and obtain the joint weighted polarization difference after summing, and construct a difference image. The joint weighted polarization difference is: $\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>$ Among them, a and b are weighting coefficients and satisfy a+b=1, a and b determine the contribution of the two difference types to the overall difference; (5) Perform threshold segmentation on the difference image to extract the change area: first, use the change detection feature quantity, that is, the mean and standard deviation of the difference image to determine the candidate threshold, and then determine the final segmentation threshold according to the ratio of the number of change points to the number of all pixels , and finally use the threshold segmentation method to extract the changed region.