A multi-graph cross-identification based background interference suppression method and system

Abstract

The application discloses a background interference suppression method and system based on multi-image cross identification. The method collects multiple original images of the same object to be identified under different imaging conditions, binds imaging condition labels, and constructs a multi-image disturbance observation set. The method performs regional decomposition on each image, establishes cross corresponding relationship between candidate region units, and generates cross-image region track chains. According to the spatial drift, morphological variation, texture response fluctuation and region response change of the cross-image region track chains, the method calculates a background interference migratory index, and screens suspected background interference track chains. According to the stable target candidate track chain, the method extracts a target invariant kernel, establishes an interference repulsion relationship, calculates a reverse interference suppression strength, generates a hierarchical suppression strategy, and differentially suppresses the suspected background interference region. The method combines the target invariant kernel consistency verification to output a background interference suppression result. The application can improve the accuracy and stability of target identification, detection or segmentation under complex background.

Description

Technical Field

[0001] This invention relates to the fields of image processing and computer vision technology, and in particular to a background interference suppression method and system based on multi-image cross recognition. Background Technology

[0002] With the development of image processing and computer vision technologies, target recognition, target detection, and target segmentation have been widely applied in scenarios such as industrial inspection, intelligent monitoring, medical image analysis, and robot vision. In actual image acquisition, the object to be identified is often affected by factors such as complex textured backgrounds, shadows, reflections, occlusions, false edges, and changes in lighting, leading to confusion between the target area and the background area, and consequently, problems such as false detection, missed detection, boundary breaks, or inaccurate segmentation.

[0003] Existing background interference suppression methods typically employ image enhancement, thresholding, edge detection, saliency analysis, attention weighting, and multi-image fusion. While these methods can improve image quality or enhance target features to some extent, most still rely primarily on single-image features or simple multi-image fusion results. They lack specific analysis of the migration, expansion, breakage, disappearance, and abrupt response changes of background interference under different imaging conditions, making it easy to misidentify background textures, reflective edges, or shadow areas as target structures.

[0004] Therefore, existing technologies still suffer from problems such as inaccurate background interference recognition, insufficient target boundary protection, and difficulty in adaptively adjusting suppression intensity in complex background scenarios. Especially when multiple images are involved in processing, how to accurately distinguish between the target subject and background interference by utilizing the differences between multiple images—where the target is relatively stable and the background interference is easily variable—and how to suppress the background while avoiding damage to target details, remains a technical problem that needs to be solved. Summary of the Invention

[0005] This invention provides a background interference suppression method based on multi-image cross-recognition, comprising: S10. Collect multiple original images of the same object to be identified under different imaging conditions, and bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set; S20. Perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence between candidate region units in different images based on multi-image cross-identification to generate cross-image region trajectory chains. S30. Based on the spatial drift, morphological variation, texture response fluctuation and regional response change of the cross-map region trajectory chain under different imaging condition labels, calculate the background interference transferability index, and screen suspected background interference trajectory chains based on the background interference transferability index. S40. Extract the target invariant kernel from the stable target candidate trajectory chain in the cross-map region trajectory chain, and establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel. S50. Calculate the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship, and the consistency of interference projection direction of the suspected background interference trajectory chain, and generate a hierarchical suppression strategy based on the reverse interference suppression strength. S60. Perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, and perform consistency verification on the suppression results based on the target invariant kernel, and output the background interference suppression results.

[0006] As described above, a background interference suppression method based on multi-image cross-recognition involves acquiring multiple original images of the same object to be identified under different imaging conditions, and binding corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set, including: S101. Acquire multiple original images of the same object to be identified under at least one imaging condition in different viewpoints, different exposures, different focal lengths, different lighting, different time frames, and different imaging channels through an image acquisition device; S102. Generate corresponding imaging condition labels based on the acquisition parameters of each original image. The imaging condition labels include at least one of the following: viewing angle change information, exposure change information, focal length change information, illumination change information, acquisition time information, and imaging channel information. S103. Perform scale unification, coarse coordinate alignment, brightness standardization and basic noise filtering on multiple original images, and store the processed image content, imaging condition labels and unified coordinate reference together to form a multi-image perturbation observation set.

[0007] As described above, a background interference suppression method based on multi-image cross-recognition includes performing region decomposition on each image in the multi-image perturbation observation set, establishing cross-correspondence relationships between candidate region units in different images based on multi-image cross-recognition, and generating cross-image region trajectory chains, including: S201. Perform region decomposition on each image in the multi-image perturbation observation set to obtain multiple candidate region units. The candidate region units are determined based on at least one of edge closure structure, texture continuity range, color consistent region, local salient region and semantic candidate region. S202. Extract the regional location information, contour morphology information, texture response information, edge direction information and semantic response information of each candidate region unit to form region cross recognition features; S203. Based on the cross-region recognition features, perform cross-image matching on candidate region units in different images, establish cross-correspondence relationships between candidate region units, and concatenate candidate region units with continuous correspondence relationships to generate cross-image region trajectory chains.

[0008] As described above, a background interference suppression method based on multi-image cross-recognition includes calculating a background interference transferability index based on the spatial drift, morphological variation, texture response fluctuation, and regional response change of cross-image region trajectory chains under different imaging condition labels, and screening suspected background interference trajectory chains based on the background interference transferability index, including: S301. For each cross-map region trajectory chain, determine the spatial drift amount based on the position changes of each candidate region unit in the cross-map region trajectory chain under a unified coordinate reference. S302. Based on the differences in contour morphology, area change, edge closure change, and texture response of each candidate region unit in the cross-map region trajectory chain, determine the morphological variation, texture response fluctuation, and region response change. S303. Based on the spatial drift, morphological variation, texture response fluctuation, regional response change, and the correlation strength between the cross-map region trajectory chain and the imaging condition label, calculate the background interference migration index, and screen the cross-map region trajectory chains whose background interference migration index meets the preset interference migration conditions as suspected background interference trajectory chains.

[0009] The background interference suppression method based on multi-graph cross-identification, as described above, includes extracting target invariant kernels from stable target candidate trajectory chains in cross-graph region trajectory chains and establishing interference repulsion relationships between suspected background interference trajectory chains and target invariant kernels, including: S401. Based on the edge closure stability, semantic continuous response strength, spatial position stability and structural continuity of the stable target candidate trajectory chain, perform consistency verification on the stable target candidate trajectory chain to obtain the target invariant kernel to construct the trajectory chain; S402. The regions in the candidate trajectory chain of stable targets that maintain spatial stability, structural continuity, edge closure and consistent semantic response under different imaging conditions are fused to obtain the target invariant kernel; S403. Based on the spatial adjacency relationship, boundary interpenetration relationship, texture conflict relationship, semantic deviation relationship and trajectory offset relationship between the suspected background interference trajectory chain and the target invariant kernel, establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel.

[0010] As described above, a background interference suppression method based on multi-image cross-identification includes calculating the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship, and the consistency of interference projection direction of suspected background interference trajectory chains, and generating a hierarchical suppression strategy based on the reverse interference suppression strength, including: S501. Based on the changing direction of the suspected background interference trajectory chain under different imaging condition labels, determine the consistency of the interference projection direction of the suspected background interference trajectory chain relative to the target invariant kernel. S502. Calculate the reverse interference suppression intensity corresponding to the suspected background interference trajectory chain based on the background interference mobility index, interference repulsion relationship, interference projection direction consistency and target boundary protection factor. S503. Based on the reverse interference suppression intensity, the suspected background interference trajectory chain is classified into levels, and a graded suppression strategy including high-intensity suppression strategy, medium-intensity suppression strategy, boundary protection suppression strategy and low-intensity suppression strategy is generated.

[0011] As described above, a background interference suppression method based on multi-image cross-recognition includes: differentially suppressing image regions corresponding to suspected background interference trajectory chains according to a hierarchical suppression strategy; verifying the consistency of the suppression results based on a target invariant kernel; and outputting background interference suppression results, including: S601. Differentiated suppression is performed on the image region corresponding to the suspected background interference trajectory chain according to the hierarchical suppression strategy. Among them, the high-intensity suppression strategy is used to weaken complex background textures, false edges and drift occlusion areas, the medium-intensity suppression strategy is used to reduce the interference response of shadow, reflection and local exposure change areas, the boundary protection suppression strategy is used to maintain the continuity of the target boundary while reducing the interference near the target boundary, and the low-intensity suppression strategy is used to perform light smoothing on areas with small local response fluctuations. S602. Based on the boundary continuity, regional integrity, and semantic response intensity of the target invariant kernel, perform consistency verification on the image after differential suppression. S603. When the target invariant kernel meets the preset consistency condition, output the background interference suppression result; when the target invariant kernel does not meet the preset consistency condition, adjust the suppression level of the corresponding image region and regenerate the background interference suppression result.

[0012] The present invention also provides a background interference suppression system based on multi-image cross-recognition, comprising: The multi-image observation construction module is used to acquire multiple original images of the same object to be identified under different imaging conditions, and to bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set. The region trajectory chain generation module is used to perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence between candidate region units in different images based on multi-image cross recognition to generate cross-image region trajectory chains. The interference migration calculation module is used to calculate the background interference migration index based on the spatial drift, morphological variation, texture response fluctuation and regional response change of cross-map region trajectory chains under different imaging condition labels, and to screen suspected background interference trajectory chains based on the background interference migration index. The invariant kernel extraction module is used to extract the target invariant kernel based on the stable target candidate trajectory chain in the cross-map region trajectory chain, and to establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel; The suppression strength calculation module is used to calculate the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship and the consistency of interference projection direction of suspected background interference trajectory chain, and to generate a graded suppression strategy based on the reverse interference suppression strength. The suppression verification output module is used to perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, and to perform consistency verification on the suppression results based on the target invariant kernel, and output the background interference suppression results.

[0013] The beneficial effects achieved by this invention are as follows: By cross-identifying multiple images of the same object under different imaging conditions, this invention generates cross-image region trajectory chains and calculates the background interference transferability index. This effectively identifies background interference regions that drift, expand, break, disappear, or exhibit abrupt responses due to changes in viewing angle, exposure, focal length, illumination, or time. Simultaneously, by extracting the target invariant kernel and establishing the interference repulsion relationship between it and the suspected background interference trajectory chains, it can protect the target boundary and main structure while suppressing interference such as complex textures, shadows, reflections, occlusions, and false edges. Furthermore, by implementing differentiated processing through reverse interference suppression intensity and hierarchical suppression strategies, and by combining target invariant kernel consistency verification to correct the suppression results, this invention improves the accuracy, stability, and reliability of target recognition, detection, or segmentation in complex backgrounds. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a flowchart of a background interference suppression method based on multi-image cross recognition provided in Embodiment 1 of this application.

[0016] Figure 2 This is a schematic diagram of a background interference suppression system based on multi-image cross recognition provided in Embodiment 2 of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1

[0019] like Figure 1 As shown, Embodiment 1 of this application provides a background interference suppression method based on multi-image cross-recognition, including the following steps: S10. Collect multiple original images of the same object to be identified under different imaging conditions, and bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set.

[0020] This step is used to acquire multi-image data with a basis for cross-comparison. While acquiring image content, the imaging conditions corresponding to each original image are recorded, giving each original image a clear imaging source, imaging state, and perturbation attributes. By binding imaging condition labels to image content, the regional variation relationships between multiple images can be organized under a unified coordinate reference, providing a basis for analyzing the spatial drift, morphological variation, texture response fluctuation, and regional response changes of candidate regions under different imaging conditions. Specifically, it includes the following sub-steps: S101. Acquire multiple original images of the same object to be identified under at least one imaging condition in different viewpoints, different exposures, different focal lengths, different lighting, different time frames, and different imaging channels through an image acquisition device.

[0021] Image acquisition equipment performs multiple image acquisitions on the same object to be identified, resulting in multiple raw images. Image acquisition equipment includes industrial cameras, ordinary webcams, multi-view cameras, microscopic imaging equipment, medical image acquisition equipment, drone image acquisition equipment, mobile terminal camera devices, and other devices with image acquisition capabilities.

[0022] Multiple original images correspond to different imaging conditions, including at least one of the following: different viewing angles, different exposures, different focal lengths, different lighting conditions, different time frames, and different imaging channels. Specifically, different viewing angles are used to record spatial relationship differences caused by changes in shooting orientation; different exposures are used to record differences in the response of reflections, highlights, shadows, and locally overexposed areas; different focal lengths are used to record changes in background blur, edge diffusion, and the sharpness of the target outline; different lighting conditions are used to record changes in shadow boundaries, sudden changes in local brightness, and reflective areas; different time frames are used to record changes in dynamic obstructions, environmental disturbances, and background movement; and different imaging channels are used to record differences in the target and background response under different spectra, depths, or sensor information.

[0023] S102. Generate corresponding imaging condition labels based on the acquisition parameters of each original image. The imaging condition labels include at least one of the following: viewing angle change information, exposure change information, focal length change information, illumination change information, acquisition time information, and imaging channel information.

[0024] Imaging condition labels are generated based on the acquisition parameters corresponding to each original image, and these labels are then bound to the corresponding original images. The imaging condition labels describe the imaging state of the original images, enabling each original image to be indexed, retrieved, and compared according to specific imaging conditions during subsequent processing.

[0025] Specifically, the viewing angle change information is used to indicate the shooting angle, shooting orientation, camera posture, or viewing angle number of the image acquisition device relative to the object to be identified; the exposure change information is used to indicate the exposure time, exposure gain, brightness compensation, or highlight response status; the focal length change information is used to indicate the focal length parameter, imaging magnification, depth of field status, or focusing distance; the illumination change information is used to indicate the light source direction, light intensity, light source position, illumination mode, or shadow formation conditions; the acquisition time information is used to indicate the acquisition timestamp, frame number, or time interval of the original image; and the imaging channel information is used to indicate the image source channel, spectral channel, depth channel, infrared channel, or other sensor channels. By establishing a correspondence between the above imaging condition labels and the image number, acquisition order, and image content of the original images, multiple original images of the same object to be identified can be organized according to imaging differences.

[0026] S103. Perform scale unification, coarse coordinate alignment, brightness standardization and basic noise filtering on multiple original images, and store the processed image content, imaging condition labels and unified coordinate reference together to form a multi-image perturbation observation set.

[0027] Basic preprocessing is performed on multiple raw images to ensure comparability in spatial scale, coordinate reference, and brightness representation. Specifically, this includes: scale unification processing to ensure that the resolution, pixel size, or target subject proportion of each image meets uniform processing requirements; coarse coordinate alignment to establish a unified coordinate reference based on the target subject position, key points, contour regions, or preset reference points in the images; brightness normalization to reduce overall brightness differences caused by different exposures, lighting conditions, and imaging channels; and basic noise filtering to reduce random noise, isolated noise points, and low-intensity noise areas generated during acquisition.

[0028] After the above processing is completed, the processed image content, corresponding imaging condition labels, and unified coordinate reference are associated and stored to form a multi-image perturbation observation set. The multi-image perturbation observation set includes processed image data, image numbers, imaging condition labels, unified coordinate reference information, and basic correspondences between images.

[0029] S20. Perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence between candidate region units in different images based on multi-image cross-identification to generate cross-image region trajectory chains.

[0030] This step is used to divide the image content in the multi-image perturbation observation set into candidate region units that can be cross-compared, and to establish cross-image correspondences based on the spatial location, structural morphology, texture response, edge features, and semantic response of the candidate region units in different images, forming a cross-image region trajectory chain to characterize the regional change process. Specifically, it includes the following sub-steps: S201. Perform region decomposition on each image in the multi-image perturbation observation set to obtain multiple candidate region units. The candidate region units are determined based on at least one of edge closure structure, texture continuity range, color consistent region, local salient region, and semantic candidate region.

[0031] Based on the images in the multi-image perturbation observation set, region decomposition processing is performed under a unified coordinate reference to obtain multiple candidate region units. Each candidate region unit is a local region in the image with continuous visual attributes, used to carry the local image information required for cross-image recognition.

[0032] Candidate region units are determined based on at least one of the following: edge closure structure, texture continuity range, color consistency region, local salient region, and semantic candidate region. Specifically, edge closure structure represents a region with continuous contour boundaries; texture continuity range represents a region with continuous texture direction or intensity; color consistency region represents a region with concentrated color distribution; local salient region represents a region with prominent brightness, edge, or texture response; and semantic candidate region represents a local region semantically related to the object to be identified. After region decomposition, each image forms a corresponding set of candidate region units.

[0033] S202. Extract the regional location information, contour morphology information, texture response information, edge direction information and semantic response information of each candidate region unit to form region cross recognition features.

[0034] Region cross-identification features are extracted for candidate regions. These features describe the local attributes required for matching candidate regions across different images. These features include region location information, contour morphology information, texture response information, edge direction information, and semantic response information. Corresponding features are selected based on different imaging conditions and region matching requirements to participate in cross-image matching and change calculation.

[0035] Among them, regional location information represents the spatial distribution of candidate region units under a unified coordinate reference; contour morphology information represents the boundary shape and regional structure of candidate region units; texture response information represents the texture variation characteristics within candidate region units; edge direction information represents the directional response at the boundary of candidate region units; and semantic response information represents the degree of semantic association between candidate region units and the object to be identified or background interference. The above information, after being correlated, forms the regional cross-identification features corresponding to the candidate region units, and provides a feature basis for cross-map region trajectory chain generation and background interference transferability index calculation.

[0036] S203. Based on the cross-region recognition features, perform cross-image matching on candidate region units in different images, establish cross-correspondence relationships between candidate region units, and concatenate candidate region units with continuous correspondence relationships to generate cross-image region trajectory chains.

[0037] Based on the cross-region recognition feature, cross-image matching is performed on candidate region units in different images to determine the cross-correspondence between candidate region units. Cross-image matching is judged based on at least one of the following: proximity of region location, similarity of contour shape, correlation of texture response, consistency of edge direction, and proximity of semantic response. The correspondence between candidate region units in different images is determined by combining imaging condition labels.

[0038] For candidate region units that form a corresponding relationship in multiple images, they are concatenated according to the image acquisition order, image number, or imaging condition label to generate a cross-image region trajectory chain. The cross-image region trajectory chain represents the continuous change relationship of candidate region units under different imaging conditions, and its recorded content includes at least one of the following: the corresponding identifier of the candidate region unit, the image information to which it belongs, the imaging condition information, and the region change information.

[0039] S30. Based on the spatial drift, morphological variation, texture response fluctuation and regional response change of the cross-map region trajectory chain under different imaging condition labels, calculate the background interference transferability index, and screen suspected background interference trajectory chains based on the background interference transferability index.

[0040] This step is used to quantify the changes in cross-map region trajectory chains. By analyzing the positional shifts, morphological changes, texture fluctuations, and response intensity changes of candidate region units under different imaging conditions, a background interference transferability index is calculated to characterize the migration features of background interference. A higher background interference transferability index indicates that the cross-map region trajectory chain more closely matches the characteristics of background textures, shadows, reflections, occlusion edges, or false edges drifting, expanding, breaking, disappearing, or experiencing abrupt responses as imaging conditions change. Specifically, it includes the following sub-steps: S301. For each cross-map region trajectory chain, determine the spatial drift amount based on the position changes of each candidate region unit in the cross-map region trajectory chain under a unified coordinate reference.

[0041] Regarding the first A cross-map region trajectory chain is used to read the regional position information of each candidate region unit in different images, and the center position, boundary range, and spatial coverage of each candidate region unit are determined based on a unified coordinate reference. Since multiple original images have been scaled and coarsely aligned, the positional changes of candidate region units in different images can be compared within the same spatial reference range.

[0042] Based on the positional changes of candidate region units under different imaging condition labels, the first... Spatial drift is the amount of spatial drift across a cross-image region trajectory chain. Spatial drift is used to characterize the spatial stability of the same visual region in multiple images; when the candidate region unit undergoes significant positional shifts with changes in viewpoint, focal length, time frame, or imaging channel, the spatial drift increases; when the candidate region unit remains relatively stable in different images, the spatial drift decreases.

[0043] S302. Based on the differences in contour morphology, area change, edge closure change, and texture response of each candidate region unit in the cross-map region trajectory chain, determine the morphological variation, texture response fluctuation, and region response change.

[0044] Based on the region cross-identification feature, for the first Structural and response variations of candidate region units in a cross-image trajectory chain are analyzed. Morphological variation is determined based on differences in contour shape, area, boundary integrity, and morphological complexity of candidate region units in different images; texture response fluctuation is determined based on variations in texture intensity, direction, density, and continuity within candidate region units; and region response variation is determined based on the strength variations of edge response, saliency response, and semantic response of candidate region units in different images.

[0045] For stable regions corresponding to the target object, their contour morphology, edge closure degree, and semantic response usually remain continuous under different imaging conditions. For interfering regions such as background texture, shadows, reflections, occluded edges, and false edges, their morphological boundaries, texture distribution, and response intensity are prone to fluctuations with changes in imaging conditions. Therefore, morphological variation, texture response fluctuation, and regional response change are used together to describe the unstable variation characteristics of cross-image region trajectory chains.

[0046] S303. Based on the spatial drift, morphological variation, texture response fluctuation, regional response change, and the correlation strength between the cross-map region trajectory chain and the imaging condition label, calculate the background interference migration index, and screen the cross-map region trajectory chains whose background interference migration index meets the preset interference migration conditions as suspected background interference trajectory chains.

[0047] The spatial drift, morphological variation, texture response fluctuation, regional response change, and imaging condition label association strength are fused together to obtain the first... Background interference transferability index for cross-map region trajectory chains. This index measures the degree of unstable migration of cross-map region trajectory chains as imaging conditions change, and provides inverse constraints on stable target structures through edge closure stability, semantic persistence response strength, and structural continuity.

[0048] When the When a cross-regional trajectory chain contains at least two candidate region units, its background disturbance transferability index is... Calculated using the following formula: ,in, Indicates the first Background disturbance transferability index for cross-regional trajectory chains; This represents a normalization mapping function used to map calculation results to a preset numerical range; Indicates the first The number of candidate region units contained in a cross-graph region trajectory chain; Indicates the first The sequence number of the candidate region unit in the cross-regional trajectory chain; Indicates the first The first cross-region trajectory chain The candidate region unit and the first +1 imaging condition difference weight between candidate region units, which is determined based on the differences in viewing angle, exposure, focal length, illumination, acquisition time or imaging channel between the corresponding images of adjacent candidate region units. Indicates the first The correlation strength between the changing trend of cross-map region trajectory chains and imaging condition labels; Indicates the first The first cross-region trajectory chain The center position vector of each candidate region unit under a unified coordinate datum; Indicates the first The average area of ​​each candidate region unit in a cross-regional trajectory chain; This represents a stable parameter used to avoid the denominator being zero; Indicates the first The first cross-region trajectory chain The boundary range of each candidate region unit; IoU( () indicates the degree of overlap between two boundary ranges; Indicates the first The first cross-region trajectory chain The texture response vector of each candidate region unit after uniform dimension encoding; Indicates the first The first cross-region trajectory chain Regional response intensity of each candidate region unit; Indicates the first The average regional response intensity of each candidate region unit in a cross-regional trajectory chain; Indicates the first Edge closure stability of a cross-graph region trajectory chain; Indicates the first The semantic sustained response strength of a cross-graph region trajectory chain; Indicates the first Structural continuity of a cross-regional trajectory chain; , , , These represent the weighting coefficients for spatial drift, boundary change, texture fluctuation, and regional response change, respectively. , , These represent the inverse constraint coefficients for edge closure stability, semantic persistence response strength, and structural continuity, respectively. When the... When a cross-regional trajectory chain contains only one candidate region unit, the cross-regional trajectory chain is treated as an isolated candidate region for single-point interference judgment and is not included in the calculation of continuous changes between adjacent candidate region units. After completing the calculation of the background interference transferability index, Compare with a preset interference migration threshold. When When the preset interference migration conditions are met, the corresponding cross-map region trajectory chain is marked as a suspected background interference trajectory chain; when If the preset interference migration conditions are not met, but the edge closure stability, semantic persistence response strength, and structural continuity meet the preset stability conditions, the corresponding cross-graph region trajectory chain is marked as a stable target candidate trajectory chain. This yields a set of suspected background interference trajectory chains and a set of stable target candidate trajectory chains.

[0049] S40. Extract the target invariant kernel from the stable target candidate trajectory chain in the cross-map region trajectory chain, and establish the interference rejection relationship between the suspected background interference trajectory chain and the target invariant kernel.

[0050] This step is used to extract the target invariant kernel based on the stable target candidate trajectory chain obtained in step S30, and to analyze the interference effect of suspected background interference trajectory chains using the target invariant kernel as a reference. The target invariant kernel refers to the stable core region of the target that maintains consistent spatial coverage, consistent edge closure, structural continuity, and consistent semantic response under different imaging conditions. It is used for subsequent interference repulsion relationship establishment, reverse interference suppression intensity calculation, and suppression result consistency verification. Specifically, it includes the following sub-steps: S401. Based on the edge closure stability, semantic continuous response strength, spatial position stability and structural continuity of the stable target candidate trajectory chain, perform consistency verification on the stable target candidate trajectory chain to obtain the target invariant kernel to construct the trajectory chain.

[0051] Based on the set of stable target candidate trajectory chains obtained in step S30, a consistency check is performed on each stable target candidate trajectory chain. For the first... Read the edge closure stability of a stable target candidate trajectory chain. Semantic sustained response strength Structural continuity and background interference transferability index Furthermore, by combining the positional distribution of candidate region units in the trajectory chain under a unified coordinate reference, it is determined whether they continuously correspond to the stable structural region of the object to be identified.

[0052] When a stable target candidate trajectory chain satisfies the following conditions: background interference transferability index is lower than a preset interference transfer threshold, edge closure stability meets a preset closure condition, semantic continuity response strength meets a preset semantic condition, and structural continuity meets a preset continuity condition, it is retained as a trajectory chain for constructing the target invariant kernel. For trajectory chains that are stable only in a few images, lack continuous correspondence, or are inconsistent with the semantic response of the target subject, they are not included in the construction of the target invariant kernel, thus obtaining the set of trajectory chains constructed by the target invariant kernel.

[0053] S402. The regions in the candidate trajectory chain of the stable target that maintain spatial stability, structural continuity, edge closure and consistent semantic response under different imaging conditions are fused to obtain the target invariant kernel.

[0054] Candidate region units in the trajectory chain set constructed by the target invariant kernel are mapped to a unified coordinate reference, and fusion processing is performed based on the spatial coverage consistency, edge closure consistency, structural connectivity consistency, and semantic response consistency of the candidate region units in different images. During fusion, the region jointly supported by multiple images is used as the target core region, and regions that only appear in a single image, lack cross-image continuous support, or have inconsistent responses with the target subject are excluded.

[0055] After fusion, a target-invariant kernel is formed. The target-invariant kernel is used to characterize the main structural region of the object to be identified that remains stable under different imaging conditions. It includes the target core region range, the stable boundary range, and the target boundary protection range. The target core region is used to represent the stable target part supported by multiple images, the stable boundary range is used to represent the target contour that remains continuous in different images, and the target boundary protection range is used to mark the transition region around the target-invariant kernel where the suppression intensity needs to be reduced.

[0056] S403. Based on the spatial adjacency relationship, boundary interpenetration relationship, texture conflict relationship, semantic deviation relationship and trajectory offset relationship between the suspected background interference trajectory chain and the target invariant kernel, establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel.

[0057] Using the target invariant core as a reference, a relative relationship analysis is performed on the suspected background interference trajectory chains obtained in step S30. For each suspected background interference trajectory chain, its corresponding candidate region unit is mapped to a unified coordinate reference, and its spatial relationship with the target core region, stable boundary range, and target boundary protection range is determined. Furthermore, it is analyzed whether its cross-map change direction is towards the target invariant core or crosses the target boundary protection range.

[0058] Based on the aforementioned relative relationships, an interference repulsion relationship is established between the suspected background interference trajectory chain and the target invariant kernel. This interference repulsion relationship includes at least one of the following: spatial adjacency, boundary interpenetration, texture conflict, semantic deviation, and trajectory offset. Specifically, spatial adjacency represents the distance and contact state between the suspected background interference trajectory chain and the target invariant kernel; boundary interpenetration represents the degree of intrusion of the suspected background interference trajectory chain into the stable boundary range; texture conflict represents the difference in texture response between the suspected background interference trajectory chain and the target core region; semantic deviation represents the degree of deviation between the suspected background interference trajectory chain and the target main semantic response; and trajectory offset represents the cross-graph offset direction and magnitude of the suspected background interference trajectory chain relative to the target invariant kernel. This yields a set of interference repulsion relationships used to calculate the reverse interference suppression intensity.

[0059] S50. Calculate the reverse interference suppression intensity based on the background interference migration index, interference repulsion relationship, and the consistency of interference projection direction of the suspected background interference trajectory chain, and generate a hierarchical suppression strategy based on the reverse interference suppression intensity.

[0060] This step quantifies the impact of suspected background interference trajectory chains relative to the target invariant kernel. By back-projecting the cross-map migration direction of the suspected background interference trajectory chains, it determines whether they are moving towards the target invariant kernel, whether they have entered the target boundary protection range, and whether they interfere with the stable boundary. Based on this, the back-interference suppression strength is calculated. Specifically, it includes the following sub-steps: S501. Based on the changing direction of the suspected background interference trajectory chain under different imaging condition labels, determine the consistency of the interference projection direction of the suspected background interference trajectory chain relative to the target invariant kernel.

[0061] Based on the positional changes of each candidate region unit in the suspected background interference trajectory chain under a unified coordinate reference, the cross-map migration direction of the suspected background interference trajectory chain is determined. The cross-map migration direction is jointly determined by the center position offset, boundary range change, and regional coverage change between adjacent candidate region units, and is used to characterize the movement, expansion, or deviation trend of the suspected background interference region under different imaging conditions.

[0062] After obtaining the cross-map migration direction, this direction is back-projected onto the unified coordinate datum where the target invariant nucleus is located, and compared with the spatial direction of the suspected background interference trajectory chain pointing towards the target invariant nucleus to obtain the consistency of the interference projection direction. The consistency of the interference projection direction is used to characterize whether the changing trend of the suspected background interference trajectory chain continues towards the target invariant nucleus, whether it enters the target boundary protection range, and whether it has an intervening influence on the stable boundary range; the higher the consistency of the direction, the greater the possibility that the suspected background interference trajectory chain will interfere with the target's main structure. The back-projection results are used to determine the degree of projection intrusion of the suspected background interference trajectory chain relative to the target invariant nucleus.

[0063] S502. Calculate the reverse interference suppression intensity corresponding to the suspected background interference trajectory chain based on the background interference mobility index, interference repulsion relationship, interference projection direction consistency, and target boundary protection factor.

[0064] For each suspected background interference trajectory chain, the reverse interference suppression strength is calculated. This reverse interference suppression strength is used to characterize the degree to which the region corresponding to the suspected background interference trajectory chain should be weakened during the background interference suppression process. Its calculation takes into account the migration capability of the background interference of the suspected background interference trajectory chain itself, the degree of repulsion between it and the target invariant core, the degree to which it enters the target boundary protection range along the back projection direction, and the degree to which the target boundary needs to be protected.

[0065] No. Inverse interference suppression strength of suspected background interference trajectory chains Calculated using the following formula: ,in, Indicates the first The reverse interference suppression strength of a suspected background interference trajectory chain; This represents the normalization mapping function, used to map the calculation results to a preset suppression intensity range; Indicates the first Background interference migration index of a suspected background interference trajectory chain; Indicates the first The interference repulsion strength between a suspected background interference trajectory chain and the target invariant kernel is determined by the spatial adjacency relationship, boundary interpenetration relationship, texture conflict relationship, semantic deviation relationship and trajectory offset relationship in step S40; Indicates the first The degree of projection intrusion of a suspected background interference trajectory chain relative to the target invariant kernel is determined by the back projection result in step S501; Indicates the first The target boundary protection factor of the region corresponding to a suspected background interference trajectory chain is determined by the distance between the region and the target boundary protection range, the structural continuity and the edge closure maintenance. , , These represent the adjustment coefficients for interference repulsion intensity, projection intrusion degree, and target boundary protection factor, respectively. Used to characterize the degree to which the trajectory chain of suspected background interference undergoes unstable migration as imaging conditions change; This is used to characterize the degree of repulsion between the suspected background interference trajectory chain and the target invariant kernel; Used to characterize the extent to which the suspected background interference trajectory chain enters the target boundary protection range or stable boundary range along the back projection direction; This is used to reduce the suppression intensity in areas adjacent to the target boundary and in areas where the target structure is continuous. When the suspected background interference trajectory chain has a high background interference mobility index, a strong interference repulsion relationship, and a high degree of projection intrusion, the reverse interference suppression intensity increases; when the region corresponding to the suspected background interference trajectory chain has structural continuity with the target invariant core or is located within the target boundary protection range, the reverse interference suppression intensity is constrained by the target boundary protection factor.

[0066] S503. Based on the reverse interference suppression intensity, the suspected background interference trajectory chain is classified into levels, and a graded suppression strategy including high-intensity suppression strategy, medium-intensity suppression strategy, boundary protection suppression strategy and low-intensity suppression strategy is generated.

[0067] The suppression intensity of the back-interference is compared with a preset suppression level threshold, and the suppression strategy corresponding to each suspected background interference trajectory chain is determined by combining the target boundary protection factor, the degree of projection intrusion, and the interference repulsion intensity. For suspected background interference trajectory chains with high back-interference suppression intensity, high interference repulsion intensity, and low target boundary protection factor, a high-intensity suppression strategy is determined; for suspected background interference trajectory chains with back-interference suppression intensity in the middle range and local projection intrusion, a medium-intensity suppression strategy is determined; for suspected background interference trajectory chains close to the target boundary protection range and with a high target boundary protection factor, a boundary protection suppression strategy is determined; and for suspected background interference trajectory chains with low back-interference suppression intensity and low projection intrusion, a low-intensity suppression strategy is determined.

[0068] A tiered suppression strategy guides subsequent background interference processing. High-intensity suppression strategies weaken background textures, drift occlusions, and false edge regions that exhibit significant cross-image migration and high repulsion from the target's invariant kernel. Medium-intensity suppression strategies reduce shadows, reflections, and localized exposure abrupt changes in areas with some projection intrusion but without significantly disrupting the target's stable boundary. Boundary-protective suppression strategies preserve the target's contour continuity and structural integrity while reducing interference near the target boundary. Low-intensity suppression strategies provide light suppression for regions with weak interference effects, far from the target's invariant kernel, or with dispersed change trends. This forms a tiered suppression strategy set corresponding to suspected background interference trajectory chains.

[0069] S60. Perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, and perform consistency verification on the suppression results based on the target invariant kernel, and output the background interference suppression results.

[0070] This step is used to differentiate the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy generated in step S50, and to verify the suppressed image results using a target-invariant kernel to ensure that while background interference is weakened, the target structure and target boundaries are not over-suppressed. Specifically, it includes the following sub-steps: S601. Differentiated suppression is performed on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy.

[0071] Based on the suppression strategy corresponding to the suspected background interference trajectory chain, differential suppression is performed on the image region or feature region where it is located. For regions applicable to high-intensity suppression strategies, their texture response, edge response, and local saliency are reduced; for regions applicable to medium-intensity suppression strategies, their brightness abrupt changes, shadow boundaries, and reflective responses are weakened; for regions applicable to boundary-protection suppression strategies, the continuity of the target contour is preserved while reducing the interference response; for regions applicable to low-intensity suppression strategies, light smoothing or weakening processing is performed.

[0072] After differential suppression is performed, an initial background interference suppressed image is generated. In this image, the interference response of regions such as complex background textures, drift occlusion, shadows, reflections, and false edges is reduced, while the target subject region and target boundary region remain relatively intact.

[0073] S602. Based on the boundary continuity, regional integrity, and semantic response intensity of the target invariant kernel, perform consistency verification on the image after differential suppression.

[0074] Using the target invariant kernel obtained in step S40 as a reference, a consistency check is performed on the initial background interference suppression image. The check includes the boundary continuity, region integrity, and semantic response intensity of the region corresponding to the target invariant kernel.

[0075] When the boundary of the target invariant kernel remains continuous, the region does not shrink abnormally, and the semantic response strength meets the preset requirements, the suppression result is determined to meet the consistency requirements; when the target invariant kernel has a broken boundary, a missing region, or a decreased semantic response, the corresponding region is determined to have over-suppression, and the region is marked as a region to be adjusted.

[0076] S603. When the target invariant kernel meets the preset consistency condition, output the background interference suppression result; when the target invariant kernel does not meet the preset consistency condition, adjust the suppression level of the corresponding image region and regenerate the background interference suppression result.

[0077] The final output is determined based on the consistency verification results. When the target invariant kernel meets the preset consistency conditions, the initial background interference suppression image is determined as the background interference suppression result and used as the input image for subsequent target recognition, target detection, or target segmentation.

[0078] When the target invariant kernel does not meet the preset consistency conditions, the suppression level of the region to be adjusted or the boundary protection processing is enhanced according to the location of the region to be adjusted and the corresponding suppression strategy, and the final background interference suppression result is regenerated.

[0079] Example 2

[0080] like Figure 2 As shown, Embodiment 2 of this application provides a background interference suppression system based on multi-image cross-recognition, comprising: The multi-image observation construction module is used to acquire multiple original images of the same object under different imaging conditions, and to bind corresponding imaging condition labels to each original image, thus constructing a multi-image perturbation observation set. It includes the following sub-modules: The image acquisition submodule is used to acquire multiple original images of the same object to be identified under at least one imaging condition in different viewpoints, different exposures, different focal lengths, different lighting, different time frames, and different imaging channels through an image acquisition device, so that the multiple original images correspond to different imaging states.

[0081] The tag binding submodule is used to generate corresponding imaging condition tags based on the acquisition parameters of each original image, and to establish a correspondence between the imaging condition tags and the image number, acquisition order and image content of the original images, so that each original image can be organized according to the differences in imaging conditions.

[0082] The observation set generation submodule is used to perform scale unification, coarse coordinate alignment, brightness standardization and basic noise filtering on multiple original images, and to associate and store the processed image content, imaging condition labels and unified coordinate reference to form a multi-image perturbation observation set.

[0083] The region trajectory chain generation module is used to perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence relationships between candidate region units in different images based on multi-image cross-identification, generating cross-image region trajectory chains. It includes the following sub-modules: The region decomposition submodule is used to decompose each image in the multi-image perturbation observation set under a unified coordinate reference to obtain multiple candidate region units. The candidate region units are determined based on at least one of the following: edge closure structure, texture continuity range, color consistent region, local salient region, and semantic candidate region.

[0084] The feature construction submodule is used to extract the regional location information, contour morphology information, texture response information, edge direction information and semantic response information of each candidate region unit to form region cross recognition features, so that the candidate region units have the feature foundation required for cross-image matching.

[0085] The trajectory chain generation submodule is used to perform cross-image matching of candidate region units in different images based on the region cross-identification features, establish cross-correspondence relationships between candidate region units, and concatenate candidate region units with continuous correspondence relationships to generate cross-image region trajectory chains.

[0086] The interference migration calculation module is used to calculate the background interference migration index based on the spatial drift, morphological variation, texture response fluctuation, and regional response change of cross-map region trajectory chains under different imaging condition labels, and to filter suspected background interference trajectory chains based on the background interference migration index. It includes the following sub-modules: The spatial drift determination submodule is used to determine the spatial drift amount for each cross-image region trajectory chain based on the position changes of each candidate region unit in the cross-image region trajectory chain under a unified coordinate reference, so as to characterize the spatial stability of the same visual region in multiple images.

[0087] The variation determination submodule is used to determine the morphological variation, texture response fluctuation, and regional response variation based on the contour morphology differences, regional area changes, edge closure changes, and texture response differences of each candidate region unit in the cross-graph region trajectory chain.

[0088] The interference index calculation submodule is used to calculate the background interference transferability index based on the spatial drift, morphological variation, texture response fluctuation, regional response change, and the correlation strength between cross-map region trajectory chains and imaging condition labels. It also filters suspected background interference trajectory chains and stable target candidate trajectory chains based on the background interference transferability index.

[0089] The invariant kernel extraction module is used to extract target invariant kernels from stable target candidate trajectory chains in cross-map region trajectory chains, and to establish interference repulsion relationships between suspected background interference trajectory chains and target invariant kernels. It includes the following sub-modules: The consistency verification submodule is used to perform consistency verification on the candidate trajectory chain of stable targets based on the edge closure stability, semantic continuous response strength, spatial position stability and structural continuity, so as to obtain the target invariant kernel to construct the trajectory chain.

[0090] The invariant kernel generation submodule is used to map candidate region units in the trajectory chain constructed by the target invariant kernel to a unified coordinate reference, and to fuse regions that maintain consistent spatial coverage, consistent edge closure, consistent structural continuity and consistent semantic response under different imaging conditions to obtain the target invariant kernel.

[0091] The exclusion relationship establishment submodule is used to establish the interference exclusion relationship between the suspected background interference trajectory chain and the target invariant kernel, with the target invariant kernel as a reference, based on the spatial adjacency relationship, boundary interpenetration relationship, texture conflict relationship, semantic deviation relationship and trajectory offset relationship between the suspected background interference trajectory chain and the target invariant kernel.

[0092] The suppression strength calculation module is used to calculate the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship, and the consistency of interference projection direction of suspected background interference trajectory chains, and to generate a graded suppression strategy based on the reverse interference suppression strength. It includes the following sub-modules: The projection direction determination submodule is used to determine the consistency of the interference projection direction of the suspected background interference trajectory chain relative to the target invariant nucleus based on the changing direction of the suspected background interference trajectory chain under different imaging condition labels, and to determine the degree of projection intrusion of the suspected background interference trajectory chain relative to the target invariant nucleus.

[0093] The suppression intensity calculation submodule is used to calculate the reverse interference suppression intensity corresponding to the suspected background interference trajectory chain based on the background interference migration index, interference repulsion relationship, interference projection direction consistency and target boundary protection factor, so as to determine the degree of background interference reduction in the corresponding image region.

[0094] The hierarchical strategy generation submodule is used to classify suspected background interference trajectory chains according to the reverse interference suppression intensity, and generate hierarchical suppression strategies including high-intensity suppression strategy, medium-intensity suppression strategy, boundary protection suppression strategy and low-intensity suppression strategy.

[0095] The suppression verification output module is used to differentially suppress image regions corresponding to suspected background interference trajectory chains according to a hierarchical suppression strategy, and to perform consistency verification on the suppression results based on the target invariant kernel, outputting the background interference suppression results. It includes the following sub-modules: The differential suppression submodule is used to perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, so as to reduce the interference response of complex background textures, drift occlusion, shadows, reflections and false edge regions.

[0096] The consistency verification submodule is used to perform consistency verification on the image after differential suppression based on the boundary continuity, regional integrity and semantic response strength of the target invariant kernel, and to mark the region to be adjusted when the target invariant kernel has boundary breaks, region missing or semantic response decrease.

[0097] The result output submodule is used to output the background interference suppression result when the target invariant kernel meets the preset consistency condition; when the target invariant kernel does not meet the preset consistency condition, it adjusts the suppression level of the corresponding image region or enhances the boundary protection processing, and regenerates the final background interference suppression result.

[0098] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A background interference suppression method based on multi-image cross-referencing, characterized in that, include: S10. Collect multiple original images of the same object to be identified under different imaging conditions, and bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set. S20. Perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence between candidate region units in different images based on multi-image cross-identification to generate cross-image region trajectory chains. S30. Based on the spatial drift, morphological variation, texture response fluctuation and regional response change of the cross-map region trajectory chain under different imaging condition labels, calculate the background interference transferability index, and screen suspected background interference trajectory chains based on the background interference transferability index. S40. Extract the target invariant kernel from the stable target candidate trajectory chain in the cross-map region trajectory chain, and establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel. S50. Calculate the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship, and the consistency of interference projection direction of the suspected background interference trajectory chain, and generate a hierarchical suppression strategy based on the reverse interference suppression strength. S60. Perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, and perform consistency verification on the suppression results based on the target invariant kernel, and output the background interference suppression results.

2. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, Acquire multiple original images of the same object under different imaging conditions, and bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set, including the following sub-steps: S101. Acquire multiple original images of the same object to be identified under at least one imaging condition in different viewpoints, different exposures, different focal lengths, different lighting, different time frames, and different imaging channels through an image acquisition device; S102. Generate corresponding imaging condition labels based on the acquisition parameters of each original image. The imaging condition labels include at least one of the following: viewing angle change information, exposure change information, focal length change information, illumination change information, acquisition time information, and imaging channel information. S103. Perform scale unification, coarse coordinate alignment, brightness standardization and basic noise filtering on multiple original images, and store the processed image content, imaging condition labels and unified coordinate reference together to form a multi-image perturbation observation set.

3. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, The process involves performing region decomposition on each image in the multi-image perturbation observation set, establishing cross-correspondence relationships between candidate region units in different images based on multi-image cross-identification, and generating cross-image region trajectory chains, including the following sub-steps: S201. Perform region decomposition on each image in the multi-image perturbation observation set to obtain multiple candidate region units. The candidate region units are determined based on at least one of edge closure structure, texture continuity range, color consistent region, local salient region and semantic candidate region. S202. Extract the regional location information, contour morphology information, texture response information, edge direction information and semantic response information of each candidate region unit to form region cross recognition features; S203. Based on the cross-region recognition features, perform cross-image matching on candidate region units in different images, establish cross-correspondence relationships between candidate region units, and concatenate candidate region units with continuous correspondence relationships to generate cross-image region trajectory chains.

4. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, Based on the spatial drift, morphological variation, texture response fluctuation, and regional response change of cross-region trajectory chains under different imaging condition labels, the background interference transferability index is calculated, and suspected background interference trajectory chains are screened based on the background interference transferability index, including the following sub-steps: S301. For each cross-map region trajectory chain, determine the spatial drift amount based on the position changes of each candidate region unit in the cross-map region trajectory chain under a unified coordinate reference. S302. Based on the differences in contour morphology, area change, edge closure change, and texture response of each candidate region unit in the cross-map region trajectory chain, determine the morphological variation, texture response fluctuation, and region response change. S303. Based on the spatial drift, morphological variation, texture response fluctuation, regional response change, and the correlation strength between the cross-map region trajectory chain and the imaging condition label, calculate the background interference migration index, and screen the cross-map region trajectory chains whose background interference migration index meets the preset interference migration conditions as suspected background interference trajectory chains.

5. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, Based on the stable target candidate trajectory chains in the cross-map region trajectory chains, target invariant kernels are extracted, and interference repulsion relationships are established between suspected background interference trajectory chains and target invariant kernels, including the following sub-steps: S401. Based on the edge closure stability, semantic continuous response strength, spatial position stability and structural continuity of the stable target candidate trajectory chain, perform consistency verification on the stable target candidate trajectory chain to obtain the target invariant kernel to construct the trajectory chain; S402. The regions in the candidate trajectory chain of stable targets that maintain spatial stability, structural continuity, edge closure and consistent semantic response under different imaging conditions are fused to obtain the target invariant kernel; S403. Based on the spatial adjacency relationship, boundary interpenetration relationship, texture conflict relationship, semantic deviation relationship and trajectory offset relationship between the suspected background interference trajectory chain and the target invariant kernel, establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel.

6. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, Based on the background interference mobility index, interference repulsion relationship, and the consistency of interference projection direction of suspected background interference trajectory chains, the reverse interference suppression strength is calculated, and a hierarchical suppression strategy is generated based on the reverse interference suppression strength, including the following sub-steps: S501. Based on the changing direction of the suspected background interference trajectory chain under different imaging condition labels, determine the consistency of the interference projection direction of the suspected background interference trajectory chain relative to the target invariant kernel. S502. Calculate the reverse interference suppression intensity corresponding to the suspected background interference trajectory chain based on the background interference mobility index, interference repulsion relationship, interference projection direction consistency and target boundary protection factor. S503. Based on the reverse interference suppression intensity, the suspected background interference trajectory chain is classified into levels, and a graded suppression strategy including high-intensity suppression strategy, medium-intensity suppression strategy, boundary protection suppression strategy and low-intensity suppression strategy is generated.

7. The background interference suppression method based on multi-image cross-recognition according to claim 1, characterized in that, The image regions corresponding to suspected background interference trajectory chains are differentially suppressed according to a hierarchical suppression strategy, and the suppression results are verified for consistency based on the target invariant kernel. The background interference suppression results are then output, including the following sub-steps: S601. Differentiated suppression is performed on the image region corresponding to the suspected background interference trajectory chain according to the hierarchical suppression strategy. Among them, the high-intensity suppression strategy is used to weaken complex background textures, false edges and drift occlusion areas, the medium-intensity suppression strategy is used to reduce the interference response of shadow, reflection and local exposure change areas, the boundary protection suppression strategy is used to maintain the continuity of the target boundary while reducing the interference near the target boundary, and the low-intensity suppression strategy is used to perform light smoothing on areas with small local response fluctuations. S602. Based on the boundary continuity, regional integrity, and semantic response intensity of the target invariant kernel, perform consistency verification on the image after differential suppression. S603. When the target invariant kernel meets the preset consistency condition, output the background interference suppression result; when the target invariant kernel does not meet the preset consistency condition, adjust the suppression level of the corresponding image region and regenerate the background interference suppression result.

8. A background interference suppression system based on multi-image cross-recognition, characterized in that, include: The multi-image observation construction module is used to acquire multiple original images of the same object to be identified under different imaging conditions, and to bind corresponding imaging condition labels to each original image to construct a multi-image perturbation observation set. The region trajectory chain generation module is used to perform region decomposition on each image in the multi-image perturbation observation set, and establish cross-correspondence between candidate region units in different images based on multi-image cross recognition to generate cross-image region trajectory chains. The interference migration calculation module is used to calculate the background interference migration index based on the spatial drift, morphological variation, texture response fluctuation and regional response change of cross-map region trajectory chains under different imaging condition labels, and to screen suspected background interference trajectory chains based on the background interference migration index. The invariant kernel extraction module is used to extract the target invariant kernel based on the stable target candidate trajectory chain in the cross-map region trajectory chain, and to establish the interference repulsion relationship between the suspected background interference trajectory chain and the target invariant kernel; The suppression strength calculation module is used to calculate the reverse interference suppression strength based on the background interference mobility index, interference repulsion relationship and the consistency of interference projection direction of suspected background interference trajectory chain, and to generate a graded suppression strategy based on the reverse interference suppression strength. The suppression verification output module is used to perform differential suppression on the image regions corresponding to suspected background interference trajectory chains according to the hierarchical suppression strategy, and to perform consistency verification on the suppression results based on the target invariant kernel, and output the background interference suppression results.

Images