CN-121982345-A - High-precision connection point matching method for heterogeneous SAR images

Abstract

The invention discloses a high-precision connection point matching method of heterogeneous SAR images, which comprises the steps of extracting object side connection points as area guidance according to object side SAR image data, determining initial positions and guide areas of the image side connection points based on object image conversion relations, searching local optimal connection points, extracting complementary feature points according to index weighted mean ratio of an image texture structure, constructing SAR image feature descriptors to match images Fang Jingque according to gradient statistical histograms, further eliminating mismatching points through similarity threshold and random sample consistency algorithm, and obtaining high-precision connection points.

Inventors

• ZHANG ZHE
• XU BING
• Su Yunce
• CAO ZHICHAO

Assignees

• 中南大学

Dates

Publication Date

20260505

Application Date

20260109

Claims (6)

1. 1. The high-precision connection point matching method for the different-track SAR image is characterized by comprising the following steps of: Step 1, preprocessing SAR image data, carrying out object space connection point matching on an overlapping region geographical system pre-orthographic image, obtaining an initial matching result, determining an image space guiding region, and carrying out denoising and filtering processing on the SAR image; step 2, converting all object side connection points to an image side according to an object-image conversion relation, obtaining initial positions of all image side connection points, searching local optimal connection points, and determining a position expression of the image side point as follows: ; ; Wherein, the Representing the rank location of geographically matched connection points, The parameters of the geographical transformation are represented, Representing the longitude and latitude of the object space connection point; Namely, an object image conversion relation, and inputting the longitude and latitude of the object space connecting point to obtain the initial position of the image space connecting point; Supplementing based on the number of image connection points, and extracting irregularly distributed feature points of the SAR image overlapping region by adopting an exponential weighted average ratio as candidates; Step 4, constructing an image feature descriptor with strong robustness to rotation, scale and radiation change based on multi-scale gradient features, and eliminating rotation distortion caused by different-track image visual angle differences by adaptively correcting the rotation main directions of feature points, so that feature matching consistency among different-source images is improved; step 5, obtaining a matching result in a correlation threshold mode according to the characteristic matching result; and 6, calculating and analyzing the quality and the precision of the matching result, and further removing the error matching points by using a random sample consistency algorithm and affine transformation to obtain high-precision connection points.
2. 2. The method for matching high-precision connection points of different-track SAR images according to claim 1, wherein said step 1 comprises the steps of: Step 1.1, performing geocoding on SAR data according to reference DEM data to obtain a conversion relation between SAR pre-orthographic intensity images and object images; Step 1.2, object space connection point matching is carried out in an overlapping region of SAR pre-orthographic intensity images, and object space connection point matching results are obtained; And 1.3, filtering the object space connection point matching result so as to facilitate the extraction and matching of the subsequent characteristic points.
3. 3. The method for matching high-precision connection points of different-track SAR images according to claim 1, wherein said step 2 comprises the steps of: step 2.1, reading an object side matching homonymous point information file, and acquiring an image side initial homonymous point according to a pre-orthographic image geographic conversion parameter and an object image conversion relation; and 2.2, extracting the characteristics of the area near the homonymous point of the image side, analyzing the condition of the characteristic point in the search window, and adjusting the position of the homonymous point.
4. 4. The method for matching high-precision connection points of different-track SAR images according to claim 1, wherein said step 6 comprises the steps of: step 6.1, calculating the coincidence precision in the connection point based on the result of the matched connection point to obtain a calculation point in coincidence with the precision; and 6.2, carrying out affine transformation on the calculation points conforming to the precision by utilizing random sampling and screening points, and eliminating connection points with larger rough differences.
5. 5. The method for matching high-precision connection points of different-track SAR images according to claim 1, wherein said step 2 further comprises: because the object-image conversion relationship may be inaccurate, local optimal points are searched in the image space, and the offset is updated at the same time; ; ; Wherein, the Is that A search window in the vicinity of the search window, All pixel sets for the search window.
6. 6. The method for matching high-precision connection points of different-track SAR images according to claim 1, wherein the formula in step S3 is: ; ; ; Wherein, the , , , Respectively represents local exponential weighted average values of any pixel point in the SAR image in the up, down, left and right directions, , The vertical and horizontal mean ratio results are shown, , The horizontal and vertical gradient calculation results of any pixel point in the SAR image are represented; Representing the magnitude of the gradient at each pixel point, Indicating the direction.

Description

High-precision connection point matching method for heterogeneous SAR images Technical Field The invention relates to the technical field of SAR/InSAR image processing, in particular to a high-precision connection point matching method for heterogeneous SAR images. Background In heterogeneous and different-track SAR image processing, high-precision connection point matching is a core premise for realizing image registration, splicing and subsequent regional network adjustment. The connection point is used as a space association anchor point between different-track SAR images, and the matching precision directly determines the consistency of the image space positions, so that the geometric precision and the reliability of topographic mapping are affected. Along with the development of SAR technology to high resolution and multi-polarization directions, the data scale and heterogeneous characteristics of the different-track SAR image are synchronously aggravated, and double severe requirements of high-density coverage and high-precision positioning are provided for connection point matching. The different-track SAR images have obvious differences in imaging view angles, track parameters, time phases and the like, so that geometric distortion, radiation differences and other 'heterogeneous' characteristics exist among the images, and great challenges are brought to accurate matching of connecting points. Different from the requirement that the product level regional network adjustment connection point is an object point location, the connection point which is usually required when the SAR image is subjected to parameter level regional network adjustment is an image point location, and parameters such as an imaging position, a baseline and the like are calculated by adjustment according to a Range Doppler-Phase (RD-P) relation. For this, the current heterogeneous SAR image connection point matching strategy mostly adopts a technical path of 'object space coding-feature matching-image space conversion', adopts a matching method based on feature operators such as SIFT, SURF and the like, or adopts a feature matching technology based on deep learning, and directly transfers an object space matching result to an image space as the input of a parameter level regional network adjustment. However, the path has obvious limitations that firstly, in the process of object image conversion, errors of a conversion model (a distance-Doppler model and an RPC model) and geometric distortion superposition caused by topography fluctuation can cause position deviation after object image matching characteristic points are converted to an image side, the position deviation cannot be directly used as reliable positions of image side connection points, secondary matching is needed in the image side, secondly, the conventional method lacks targeted optimization on radiation isomerism of different-track images, the radiation robustness of characteristic descriptors is insufficient and is difficult to adapt to image differences in complex scenes, thirdly, the object image matching points are distributed regularly, and an effective object image conversion error compensation mechanism is lacking, so that the distribution of the image side matching points is discrete, the matching success rate is low, and the coverage requirement of complex topography areas cannot be met. The method is influenced by the 'heterogeneous' characteristics of the different-track SAR images, and the problems of low feature point matching repetition rate, high matching ambiguity and the like of a feature operator are caused easily due to large radiation feature difference and low texture structure consistency of different-track images, so that the matching is generally difficult in an image side. In summary, the current method for matching the connection points of the different-track SAR images has obvious defects in coping with the problems of image isomerism, object-image conversion position deviation and the like, and is difficult to meet the application requirements of high-precision SAR images. Therefore, a high-precision connection point matching method capable of fusing object image conversion relation, accurately overcoming heterogeneous image interference and dynamically correcting object-image position deviation is developed, specifically, in the SAR image side matching process, the object side is used for controlling and guiding an image side matching area range, interaction scattering characteristics of a radar image and ground objects are fully considered, and algorithm matching characteristics with robustness to the SAR image are adopted. In feature expression, considering rotation distortion caused by multi-angle SAR image visual angle difference and geometric distortion caused by SAR side view imaging, constructing descriptors which are robust to scale and rotation variation, adopting a multi-scale geometric model in point location selection, extr