CN-122024088-A - SAR image change detection method integrating gravitational field and difference map

CN122024088ACN 122024088 ACN122024088 ACN 122024088ACN-122024088-A

Abstract

The invention discloses a SAR image change detection method for fusing a gravitational field and a difference image, which comprises the following steps of obtaining a double-time-phase SAR image of the same area, carrying out bilateral filtering pretreatment to obtain a denoised image, establishing image gray level and physical quality mapping, extracting a structure candidate area by utilizing a high-order gravitational field model with cube-level distance attenuation, introducing geometric anisotropy to screen out pseudo-angular point noise, generating a physical structure mask by self-adaptive matching and multi-scale morphological expansion, calculating a logarithmic ratio difference image, generating an enhanced difference image and a statistical significance mask by local texture statistical factor modulation, carrying out nonlinear multiplicative fusion on the difference image by taking the physical and statistical double masks as gating factors, and finally obtaining a final result by adopting a self-adaptive threshold segmentation algorithm with closed loop feedback. The invention effectively solves the dependence on the difference graph, effectively fills the detection hole while obviously reducing the false alarm rate, and improves the detection precision and the integrity.

Inventors

ZHAO ZHIHONG
WANG HAOYU
TONG AIHUA
YU XINRU
ZHANG RAN

Assignees

石家庄铁道大学

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (7)

1. The SAR image change detection method integrating the gravitational field and the difference map is characterized by comprising the following steps of: s1, acquiring a double-time-phase SAR image And Preprocessing an image by adopting bilateral filtering to obtain a processed double-phase SAR image And ; S2, mapping the pixel gray level of the double-time-phase SAR image into physical quality, and calculating the gravitational field modulus value based on the cubic distance attenuation factor Extracting a structure candidate region; s3, calculating the local gradient direction consistency weight and combining the local gradient direction consistency weight with the gravitational field modulus value Nonlinear coupling is carried out, and pseudo-angular point noise in the structure candidate region is removed by utilizing geometric anisotropy, so that a high-confidence structural feature map is obtained ; S4, from the double-phase structure characteristic diagram with high confidence coefficient Extracting key points, matching based on self-adaptive matching radius to identify variation structural feature points, and generating a physical structure mask by utilizing multi-scale morphological operation ; S5, utilizing the preprocessed double-phase SAR image And Calculating a logarithmic ratio difference map, and generating an enhanced difference map through local texture statistical difference factor modulation And build a statistically significant mask ; S6, utilizing the mask of the physical structure And statistical significance mask As a double gating factor, performing multiplicative fusion on the enhanced difference map to obtain a fused difference map ; S7, adopting an adaptive threshold segmentation algorithm with closed loop feedback to fuse the difference graphs And (5) dividing to obtain a change detection result.
2. The SAR image change detection method of claim 1, wherein in S2, the pixel gray scale of the SAR image is mapped to a physical mass, and the gravitational field modulus is calculated, comprising the following steps: (1) Firstly, establishing a mapping relation between a digital image space and a physical gravitational field space, wherein each pixel point in the SAR image is defined as a particle with physical quality; adopting a linear normalization strategy to pre-process the image Middle coordinates Pixel gray value at Mapping to relative physical masses within the [0,1] interval The calculation formula is as follows: ; Wherein, the The original gray value of the pixel point in the image is obtained; is the minimum value of gray values in the whole image; is the maximum value of gray values in the whole image; By this mapping, the brightest strong scatterer in the image has its relative mass Approaching 1, and the darkest background area has a relative mass Approaching 0; (2) Constructing a short Cheng Jiang interaction gravitational field, namely performing high-order correction on the gravitational model, changing the gravitational model into a square distance attenuation factor, and constructing an anti-interference short Cheng Jiang interaction field; for any pixel in the image In its defined local neighborhood In, calculate the gravitational resultant force component to which it is subjected And The calculation formula is as follows: ; ; Wherein, the For the currently calculated center pixel, its coordinates are ; As the center pixel Any one of the neighborhood pixels in the neighborhood has the coordinates of ; Is the product of the relative masses of the center pixel and the neighborhood pixels; is a cubic distance attenuation term; is a universal gravitation constant and is used for adjusting the magnitude order; (3) Calculating the gravitational field modulus and structure candidate region extraction, namely vector synthesis is carried out on the gravitational components in the two orthogonal directions, and pixels are calculated Gravitational field modulus at The calculation formula is as follows: 。
3. The SAR image change detection method of claim 1, wherein in S3, the local gradient direction consistency weight is calculated and nonlinear coupling is performed, and the method specifically comprises the following steps: First, the gradient direction angle of each pixel is calculated by Sobel operator Subsequently, the standard deviation of the gradient direction in the local window is counted And construct the direction consistency weight accordingly : ; Wherein, the For the direction consistency weight, the larger the value is, the more regular the direction is; The statistical standard deviation of the gradient direction in the local window reflects the discrete degree of the direction; finally, nonlinear coupling is carried out on the physical gravitation modulus value and the geometric direction weight to generate a high-confidence structural feature map The characteristic map is at any pixel Response value at The calculation formula is as follows: ; Wherein, the The characteristic response value is a high-confidence structural characteristic response value under double constraint; representing the physical structural strength of the pixel point for the gravitational field modulus; for the direction consistency weight, the larger the value is, the more regular the direction is; based on the statistical difference that the edge of the real structure has anisotropy and the speckle noise has isotropy, the direction consistency weight is introduced As the basis of the true and false discrimination, even if the noise point intensity is large, as long as the direction is disordered, Will approach 0 and thus calculate Removing the pseudo-corner noise, and finally obtaining a high-confidence structural feature map with the pseudo-corner noise removed by traversing the whole map calculation 。
4. The SAR image change detection method of claim 1, wherein S4 specifically comprises the steps of: (1) Dynamically calculating self-adaptive matching radius according to image corner density Dynamically adjusting matching radius The calculation formula is as follows: ; Wherein, the The ratio of the number of the feature points to the total number of pixels of the image; Limiting the radius between [3, 8] pixels as a truncation function; (2) Firstly, respectively extracting local maximum value points on a structural response diagram of a first time phase and a second time phase to construct a key point set And And then, utilize Carrying out bidirectional nearest neighbor search, and identifying unmatched points as variant structure points, wherein the calculation formula is as follows: ; Wherein, the A final identified variable structure point set; A local maximum key point set extracted from the first time phase structure feature diagram; a local maximum key point set extracted from the second time phase structure characteristic diagram; Is the key point And key point Euclidean distance between; Is taken as a point To a collection The minimum of the distances of all points in (a); Is an adaptive matching radius; (3) Constructing a multi-scale physical structure mask aiming at Each of the variation points in (a) Simulating radiation attenuation characteristics of physical fields respectively in a radial sequence Carrying out multi-layer morphological expansion and giving different weights to construct a continuous two-dimensional structural field, wherein the calculation formula is as follows: ; Wherein, the Is any pixel coordinate in the image; attenuation weights for layers with different radii; as an indication function.
5. The SAR image change detection method according to claim 1, wherein in S5, the enhanced difference map is generated by modulating the local texture statistical difference factor, and the method specifically comprises the following steps: Firstly, calculating basic logarithmic ratio difference diagram, adopting logarithmic ratio operator to convert multiplicative noise into additive noise, introducing small constant for preventing zero-removing error The calculation formula is as follows: ; Wherein, the And Respectively two-phase image in pixel The gray value at which the color is to be changed, A minor constant to prevent denominator 0; Secondly, constructing local statistical difference factors, and calculating pixels by utilizing sliding windows Local mean in neighborhood Sum of local standard deviation Respectively calculating normalized mean differences Sum of variance difference The calculation formula is as follows: ; Thirdly, generating an enhanced difference map, and multiplying the pixel level logarithmic ratio by utilizing a local statistical difference factor, wherein the calculation formula is as follows: ; Wherein, the For enhancement coefficients; finally, generating a statistical significance mask, adopting self-adaptive Sigmoid mapping based on global statistical characteristics, firstly calculating Global mean of (2) And standard deviation Determining an adaptive statistical threshold The calculation formula is as follows: ; Wherein, the Is a statistical sensitivity coefficient; generating statistical significance masks using Sigmoid function The calculation formula is as follows: ; Wherein, the Mapping the difference map into a significance probability field of an interval of [0,1 ].
6. The SAR image change detection method of claim 1, wherein the step S6 of performing multiplicative fusion by using double gating factors, comprises the following steps: nonlinear multiplicative modulation fusion model based on double gating and using physical structure mask And statistical significance mask Modulating the difference map, wherein the calculation formula is as follows: ; Wherein, the The final fused difference graph; , The gain factor is modulated, a dual gating multiplicative mechanism is adopted, namely, the gain term is 1 in a background area and keeps low noise level, and the gain term is obviously larger than 1 in a change area, so that a target signal is directionally amplified.
7. The SAR image change detection method of claim 1, wherein the adaptive threshold segmentation algorithm with closed loop feedback is adopted in S7, and the method specifically comprises the following steps: generating a final change detection result based on a closed-loop feedback control loop with a priori change rate; First, an initial threshold is calculated, and Otsu pair fusion difference map is used Performing global calculation to obtain initial segmentation threshold ; Secondly, closed loop feedback detection is carried out, and the current threshold value is set The system automatically calculates the proportion of pixels that vary below the threshold The calculation formula is as follows: ; Wherein, the The final fused difference graph; And The height and width of the image, respectively; Is an indication function, and is 1 when the condition is satisfied, otherwise, is 0; Finally, threshold adaptive iteration is carried out, feedback control logic is established, and a safety change rate threshold is preset If (1) Judging that the current threshold is valid, and enabling the final threshold to be If (1) The false alarm burst is judged, namely a large amount of backgrounds are misjudged as changes, and the algorithm automatically takes the step length Iteratively increasing the threshold, i.e And re-executing the step 2 calculation Until the change rate returns to the set range to obtain the optimal threshold value ; Generating a final change detection result by using the finally determined optimal threshold value Performing binarization segmentation on the fusion difference graph, and performing 3×3 morphological open operation to remove isolated noise points to obtain a final change detection result graph The calculation formula is as follows: ; Wherein, the Is a pixel 1 Represents a change, 0 represents no change; Representing morphological open operation.

Description

SAR image change detection method integrating gravitational field and difference map Technical Field The invention relates to the technical field of radar remote sensing image processing, in particular to a SAR image change detection method integrating a gravitational field and a difference map. Background With the rapid development of remote sensing technology, the synthetic aperture radar (SYNTHETIC APERTURE RADAR, SAR) plays an irreplaceable role in the fields of natural disaster assessment, environmental monitoring, military reconnaissance, urban planning and the like by virtue of the full-day all-weather imaging capability and strong penetrability to the physical characteristics of the earth surface. SAR image change detection aims at identifying and extracting a region where the earth surface changes by analyzing two or more SAR images acquired at different times in the same geographic region, and is one of core technologies of remote sensing application. However, under the influence of SAR imaging mechanisms, multiplicative speckle noise inevitably exists in an image, and is sensitive to imaging geometric errors, resulting in difficulty in realizing high-precision change detection. Although the existing SAR image change detection method is numerous, the method still has remarkable limitations when processing complex scenes and strong noise interference, firstly the method based on a pixel level difference image mainly depends on gray level difference, high-intensity speckle noise and a real target are difficult to distinguish in a physical layer, so that the false alarm rate of a detection result is extremely high, secondly the feature extraction method based on the difference image generally follows a single-flow serial dependence architecture, feature extraction completely depends on the quality of an initial difference image, once the initial stage is polluted by noise, errors are amplified in a cascade manner and are irreversible in a subsequent step, so that serious missed detection or false alarm is caused, in addition, the prior art is sensitive to registration errors, edge false alarm is easy to generate, and due to lack of integral perception, a detection cavity is often generated in the whole target, so that the detection integrity is seriously influenced. Disclosure of Invention The technical problem to be solved by the invention is how to provide the SAR image change detection method which can remarkably improve the robustness of an algorithm, greatly reduce the false alarm rate of change detection and ensure the fullness and complete contour of the interior of a change area. In order to solve the technical problems, the technical scheme adopted by the invention is that the SAR image change detection method integrating the gravitational field and the difference map comprises the following steps: s1, acquiring a double-time-phase SAR image AndPreprocessing an image by adopting bilateral filtering to obtain a processed SAR imageAnd; S2, mapping SAR image pixel gray scale to physical quality, calculating gravitational field modulus value based on cubic distance attenuation factorExtracting a structure candidate region; s3, calculating the local gradient direction consistency weight and combining the local gradient direction consistency weight with the gravitational field modulus value Nonlinear coupling, removing pseudo-corner noise in the structure candidate region by utilizing geometric anisotropy to obtain a high-confidence structural feature map; S4, from the double-phase structure characteristic diagram with high confidence coefficientExtracting key points, matching based on self-adaptive matching radius to identify variation structural feature points, and generating a physical structure mask by utilizing multi-scale morphological operation; S5, utilizing the preprocessed SAR imageAndCalculating a logarithmic ratio difference map, and generating an enhanced difference map through local texture statistical difference factor modulationAnd build a statistically significant mask; S6, utilizing the mask of the physical structureAnd statistical significance maskAs a double gating factor, performing multiplicative fusion on the enhanced difference map to obtain a fused difference map; And S7, dividing the fusion difference graph by adopting a self-adaptive threshold segmentation algorithm with closed loop feedback to obtain a change detection result. The physical-statistical double-flow parallel decoupling architecture constructed by the invention directly extracts the characteristics from the original image through the physical structure flow independent of the generation process of the difference image, and cuts off the cascade propagation path of the noise from the source by utilizing the heterogeneous complementary characteristics of the physical characteristics and the statistical characteristics, so that the high reliability of detection can be maintained under the condition of extremely