An image processing method, apparatus, storage medium, and device

Abstract

The application discloses an image processing method and device, a storage medium and equipment, relates to the technical field of image processing, and mainly aims to solve the problem of poor image processing accuracy of an existing aircraft. The method comprises the following steps: acquiring target image data to be processed; dividing the target image data into multiple sub-section image data according to a cross-track direction, and determining the inter-frame displacement of the standardized multiple sub-section image data according to different push-broom scenarios; fusing the multiple sub-section image data based on the inter-frame displacement to obtain spliced image data in different image formats; performing first gradual fusion processing on three-channel image data in the spliced image data to obtain target color image data, and performing second gradual fusion processing on multiple spliced image data to obtain target wide-width image data.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image processing method, apparatus, storage medium and device. Background Technology

[0002] Spaceborne pushbroom imaging is a mainstream technology for Earth observation. When spacecraft such as satellites are in orbit, they can vertically scan the Earth's surface using linear array detectors to obtain raw L0 image data. At this time, the raw L0 image data needs to undergo radiometric and geometric correction processes to generate usable L1 images.

[0003] Currently, radiometric correction of existing L1-level image data employs traditional flat-field correction methods that rely on calibration parameters. Geometric correction involves integer pixel translation, feature point matching, and global affine or polynomial transformations to complete stitching and fusion. However, traditional flat-field correction cannot handle high-frequency stripe noise, is unsuitable for dynamic changes in on-orbit parameters, and fails to address inter-frame radiometric jumps and gaps in cross-orbit stitching, thus affecting radiometric quality. Furthermore, integer pixel translation produces geometric jagged edges, is affected by attitude jitter, and results in insufficient stitching accuracy. Global affine or polynomial transformations cannot fit local high-order nonlinear misalignments, leading to "color fringes" in color images and impacting imaging quality. Therefore, an image processing method is urgently needed to address these issues. Summary of the Invention

[0004] In view of this, this application provides an image processing method, apparatus, storage medium, and device, the main purpose of which is to solve the problem of poor accuracy in image processing acquired by existing aircraft.

[0005] According to one aspect of this application, an image processing method is provided, comprising: Acquire the target image data to be processed; The target image data is divided into multiple sub-segments according to the cross-track direction, and the inter-frame displacement of the standardized multiple sub-segments is determined according to different push-broom scenarios. Based on the inter-frame displacement, the multi-segment image data is fused to obtain stitched image data in different image formats; A first gradient fusion process is performed on the three-channel image data in the stitched image data to obtain target color image data, and a second gradient fusion process is performed on multiple stitched image data to obtain target wide-format image data.

[0006] Furthermore, after dividing the target image data into multiple sub-segments according to the cross-track direction, the method further includes: The stable column response curve of the target image data is determined as the flat field correction coefficient, and the multi-segment image data is denoised based on the flat field correction coefficient and the segmentation of the multi-segment image data. The denoised multi-segment image data is linearly mapped according to a preset brightness range to obtain standardized multi-segment image data. The standardized multi-segment image data includes multi-segment image data in a first format and target image data in a second format.

[0007] Furthermore, determining the inter-frame shift of the standardized multi-segment image data according to different pushbroom scenarios includes: When the push-broom scenario is a uniform push-broom scenario, the inter-frame displacement of the standardized multi-segment image data is calculated based on the ground resolution, camera frame rate and aircraft speed. When the push-broom scenario is a non-uniform push-broom scenario, the feature points of the standardized multi-segment image data are obtained, and the inter-frame displacement is determined based on the feature points.

[0008] Furthermore, the method also includes: When the inter-frame displacement is greater than a preset threshold, the inter-frame displacement corresponding to the previous frame is determined as the inter-frame displacement of the current frame.

[0009] Furthermore, the process of fusing the multi-segment image data based on the inter-frame shift to obtain stitched image data in different image formats includes: The inter-frame displacement is determined based on the sub-pixel stitching function, which corresponds to the cumulative displacement of the multi-segment image data. The bit depth is then determined based on the coordinates corresponding to the cumulative displacement, and the weights of neighboring pixels are determined accordingly. Based on the weights of neighboring pixels and the coordinates, pixel image data is obtained by re-sampling and accumulating. Then, the pixel image data is normalized based on bilinear interpolation weights and fusion weights to obtain stitched image data in a first format and stitched image data in a second format.

[0010] Further, the first gradient fusion processing based on the three-channel image data in the stitched image data to obtain the target color image data includes: Extract the three-channel image data of the stitched image data in the first format, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; Based on the preset strip height and preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated. Weighted fusion is performed based on the local transform field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data.

[0011] Furthermore, the second gradient fusion process based on multiple stitched image data to obtain the target wide-format image data includes: Extract multiple image data of preset length from the stitched image data of the second format, and perform coarse matching on the multiple image data of preset length based on the alignment method to obtain the corrected long strip image data; The overlapping regions of the elongated image data are fused to obtain reference image data; Radiometric correction is performed on the reference image data based on the regional pixel mean to obtain the target wide-area image data.

[0012] According to another aspect of this application, an image processing apparatus is provided, comprising: The acquisition module is used to acquire the target image data to be processed; The determination module is used to divide the target image data into multiple sub-segments according to the cross-track direction, and determine the inter-frame displacement of the standardized multiple sub-segments according to different push-broom scenarios; The fusion module is used to fuse the multi-segment image data based on the inter-frame displacement to obtain stitched image data in different image formats; The processing module is used to perform a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data, and to perform a second gradient fusion process based on multiple stitched image data to obtain target wide-format image data.

[0013] Furthermore, the device also includes: The denoising module is used to determine the flat field correction coefficient based on the stable column response curve of the target image data, and to denoise the multi-segment image data based on the flat field correction coefficient and the segmentation of the multi-segment image data. The mapping module is used to perform linear mapping on the denoised multi-segment image data according to a preset brightness range to obtain standardized multi-segment image data. The standardized multi-segment image data includes multi-segment image data in a first format and target image data in a second format.

[0014] Furthermore, The determining module calculates the inter-frame displacement of the standardized multi-segment image data based on ground resolution, camera frame rate, and aircraft speed when the push-broom scenario is a uniform push-broom scenario; when the push-broom scenario is a non-uniform push-broom scenario, it acquires feature points of the standardized multi-segment image data and determines the inter-frame displacement based on the feature points.

[0015] Furthermore, the determining module is also used to determine the inter-frame displacement corresponding to the previous frame as the inter-frame displacement of the current frame when the inter-frame displacement is greater than a preset threshold.

[0016] Furthermore, The fusion module is specifically used to determine the cumulative displacement of the multi-segment image data corresponding to the inter-frame displacement based on the sub-pixel stitching function, and to perform bit division based on the coordinates corresponding to the cumulative displacement to determine the weights of neighboring pixels; to perform re-sampling accumulation based on the weights of neighboring pixels and the coordinates to obtain pixel image data, and to perform normalization processing on the pixel image data based on bilinear interpolation weights and fusion weights to obtain stitched image data in a first format and stitched image data in a second format.

[0017] Further, the fusion module is specifically used to extract three-channel image data from the stitched image data of the first format, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; based on a preset strip height and a preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated; weighted fusion is performed based on the local transformation field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data.

[0018] Furthermore, the fusion module is specifically used to extract multiple image data of preset length from the stitched image data of the second format, and to perform coarse matching on the multiple image data of preset length based on the alignment method to obtain corrected long strip image data; to perform overlapping region fusion on the long strip image data to obtain reference image data; and to perform radiometric correction on the reference image data based on the regional pixel mean to obtain the target wide image data.

[0019] According to another aspect of this application, a storage medium is provided that stores at least one executable instruction, which causes a processor to perform operations corresponding to the image processing method described above.

[0020] According to another aspect of this application, a terminal is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above image processing method.

[0021] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides an image processing method, apparatus, storage medium, and device. Compared with the prior art, the embodiments of this application acquire target image data to be processed; divide the target image data into multiple sub-segments according to the cross-track direction, and determine the inter-frame displacement of the standardized multi-segment image data according to different push-broom scenarios; fuse the multi-segment image data based on the inter-frame displacement to obtain stitched image data of different image formats; perform a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data, and perform a second gradient fusion process based on multiple stitched image data to obtain target wide-width image data, thereby achieving the purpose of seamless cross-track stitching, improving stitching accuracy, ensuring a color-edge-free image effect, and thus improving image processing accuracy.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of an image processing method provided in an embodiment of this application is shown; Figure 2 This illustration shows a segmentation diagram provided in an embodiment of this application; Figure 3 This illustration shows a schematic diagram of a satellite image processing flow provided in an embodiment of this application; Figure 4 This illustration shows a block diagram of an image processing apparatus provided in an embodiment of this application; Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of this application is shown. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] This application provides an image processing method, such as... Figure 1 As shown, the method includes: 101. Obtain the target image data to be processed.

[0027] In this embodiment, the target image data to be processed is image data acquired by spacecraft through spaceborne pushbroom scanning of a designated area on Earth. This data may include, but is not limited to, images of urban roads, forests, and oceans in raw or bin format. This embodiment does not impose specific limitations. Spaceborne pushbroom refers to pushbroom scanning imaging, where a linear array detector perpendicular to the flight direction flies forward with the satellite, scanning the ground line by line to acquire and stitch together a two-dimensional image.

[0028] It should be noted that the spacecraft in the embodiments of this application include, but are not limited to, optical remote sensing satellites, exploration satellites, etc. The current execution end can be a satellite processing end or an Earth processing end. The embodiments of this application do not make specific limitations.

[0029] 102. Divide the target image data into multiple sub-segments according to the cross-track direction, and determine the inter-frame displacement of the standardized multiple sub-segments according to different push-broom scenarios.

[0030] In this embodiment of the application, in order to eliminate the non-uniformity of the sensor during push-broom operation, the current execution end divides the target image data based on the trans-orbit direction to obtain multi-segment image data. Here, the trans-orbit direction refers to the direction perpendicular to the spacecraft's flight path and spanning both sides of the orbit, such as... Figure 2 As shown, the number of divisions can be set based on the noise reduction requirements, and this embodiment does not impose specific limitations.

[0031] It should be noted that different velocity sweeping scenarios exist during spacecraft push-brooming. Therefore, the inter-frame displacement of the standardized multi-segment image data is determined according to different sweeping scenarios. At this time, the sweeping scenario can include uniform sweeping scenario and non-uniform sweeping scenario. Standardization refers to the steps of denoising and correcting the multi-segment image data to determine the inter-frame displacement. Among them, the inter-frame displacement refers to the positional offset of ground objects between two consecutive frames or two lines of images. It can be determined by matching feature points, sub-pixels, or phase, and this application embodiment does not make specific limitations.

[0032] In another embodiment of this application, for further definition and explanation, the step of determining the inter-frame shift of the standardized multi-segment image data according to different pushbroom scenarios includes: When the push-broom scenario is a uniform push-broom scenario, the inter-frame displacement of the standardized multi-segment image data is calculated based on the ground resolution, camera frame rate and aircraft speed. When the push-broom scenario is a non-uniform push-broom scenario, the feature points of the standardized multi-segment image data are obtained, and the inter-frame displacement is determined based on the feature points.

[0033] To meet the inter-frame displacement calculation requirements under different push-broom scenarios and ensure the accuracy of inter-frame displacement calculation, thereby improving the processing efficiency of image data, the current execution terminal first determines the push-broom scenario when calculating the inter-frame displacement of multi-segment image data.

[0034] In a specific embodiment, when the push-broom scenario is a uniform push-broom scenario, the current execution terminal calculates the inter-frame displacement of the standardized multi-segment image data based on the ground resolution, camera frame rate, and aircraft speed. In this case, it can be directly expressed by the formula: ; in, This is the inter-frame offset. The speed of an aircraft, such as a satellite, is measured in m / s. This refers to the camera frame rate, measured in Hz. This represents the camera's ground resolution, measured in m / pixel.

[0035] In a specific embodiment, when the pushbroom scenario is a non-uniform pushbroom scenario, for example, when the satellite has a certain degree of jitter that gives the image rich texture, the current execution end can use methods such as feature point matching, sub-pixel matching, or phase estimation for calculation. Specifically, firstly, the feature points of the standardized multi-segment image data are obtained. At this time, the feature points are found by the Scale-Invariant Feature Transform (SIFT) matching algorithm, Oriented Fast and Rotated BRIEF (ORB), or Speeded Up Robust Features (SURF) algorithm. Stable feature points in the image are then found using K-Nearest Neighbors (knn) matching and proportional test relaxation to obtain high-confidence matching pairs. Finally, the median of the coordinate difference is used to determine the inter-frame displacement to achieve the purpose of resisting noise and incorrect matching.

[0036] In another embodiment of this application, for further definition and explanation, the steps also include: When the inter-frame displacement is greater than a preset threshold, the inter-frame displacement corresponding to the previous frame is determined as the inter-frame displacement of the current frame.

[0037] To avoid displacement jumps caused by matching errors, the current execution end employs a displacement anti-jump constraint mechanism to determine inter-frame displacement. Specifically, a preset threshold is set as the jump threshold. When the inter-frame displacement exceeds the preset threshold, the inter-frame displacement corresponding to the previous frame is determined as the inter-frame displacement of the current frame. This identifies and corrects instantaneous calculation errors caused by the algorithm's inherent instability or external interference, thereby ensuring the continuity and smoothness of the motion trajectory and greatly improving the robustness and smoothness of the entire motion trajectory estimation.

[0038] In another embodiment of this application, for further definition and explanation, after dividing the target image data into multiple sub-segments according to the cross-track direction, the method further includes: The stable column response curve of the target image data is determined as the flat field correction coefficient, and the multi-segment image data is denoised based on the flat field correction coefficient and the segmentation of the multi-segment image data. The denoised multi-segment image data is linearly mapped according to a preset brightness range to obtain standardized multi-segment image data.

[0039] To ensure the effectiveness of processing multi-segment image data, the current execution end performs standardization processing after obtaining the multi-segment image data. Specifically, firstly, flat-field correction coefficients are determined based on the stable column response curve of the target image data, so that the multi-segment image data can be denoised based on the flat-field correction coefficients and the segmentation of the multi-segment image data. The stable column response curve is generated by independently performing dual median statistics in both spatial and temporal dimensions on different strips of the target image data. The calculation formula for the stable column response curve is expressed as: ; in, For column indexes within the stripe, For stabilizing the column response curve The median is used as a reference for gain correction. To prevent extremely small positive numbers from being divided by zero, To The correction curve after applying dynamic lower limit protection can be calculated as follows: Dynamic lower limit By taking Preset lower percentile and and a preset scaling factor The larger value of the product is obtained, i.e. At this point, since the stable column response curve characterizes the photoelectric conversion characteristics of each column of pixels, the flat field correction coefficient, which characterizes the gain and bias, is obtained by fitting this curve. This coefficient is used to normalize the response of all columns to the same reference, thereby eliminating fixed-mode noise and non-uniformity of light response between columns. Therefore, the flat field correction coefficient, which avoids gain anomalies, can be determined based on the stable column response curve. Furthermore, based on the flat field correction coefficient and the segmentation of the multi-segment image data, denoising is performed on the multi-segment image data. This involves dividing the continuous gain curve into segments according to the image segments, replacing the original value of each segment with its own median value to form a stepped curve. The edges of the steps are then linearly feathered to remove stripe noise and achieve smooth transitions between segments. Finally, the denoised multi-segment image data is linearly mapped according to a preset brightness range to obtain standardized multi-segment image data. Here, the preset brightness range can be set in advance based on processing requirements, and the linear mapping method can be least squares fitting, mean normalization mapping, etc. This application embodiment does not specifically limit the method.

[0040] It should be noted that, since the target image data can initially be obtained from an independent raw or bin format file, after denoising, preferably, the 16-bit raw data can be linearly mapped according to a uniform brightness range to output an 8-bit preview image and 16-bit raw data respectively, thus completing data standardization and obtaining the standardized multi-segment image data. That is, the standardized multi-segment image data includes multi-segment image data in the first format and target image data in the second format, namely 8-bit multi-segment image data and 16-bit target image data.

[0041] In this embodiment, the pushbroom image radiometric correction and destriating method based on segmented flat field correction divides the image into multiple sub-segments along the cross-track direction, independently calculates the flat field correction gain of each sub-segment, and smooths the sub-segment boundaries through a feathering algorithm, effectively removing hard boundary stripe noise while preserving ground feature details and solving the radiometric distortion problem.

[0042] 103. Based on the inter-frame displacement, the multi-segment image data is fused to obtain stitched image data in different image formats.

[0043] In this embodiment, due to the positional shift of the standard scene objects during inter-frame displacement, a fusion method is used to accumulate the multi-segment image data to achieve the stitching purpose, thereby obtaining stitched image data in different image formats. The image formats may include stitched image data in a first format and stitched image data in a second format, preferably 8-bit and 16-bit stitched images, but this embodiment does not impose specific limitations.

[0044] In another embodiment of this application, for further definition and explanation, the step of fusing the multi-segment image data based on the inter-frame shift to obtain stitched image data of different image formats includes: The inter-frame displacement is determined based on the sub-pixel stitching function, which corresponds to the cumulative displacement of the multi-segment image data. The bit depth is then determined based on the coordinates corresponding to the cumulative displacement, and the weights of neighboring pixels are determined accordingly. Based on the weights of neighboring pixels and the coordinates, pixel image data is obtained by re-sampling and accumulating. Then, the pixel image data is normalized based on bilinear interpolation weights and fusion weights to obtain stitched image data in a first format and stitched image data in a second format.

[0045] To achieve image stitching and avoid image distortion caused by inter-frame displacement, the current execution end, when fusing multi-segment image data, first determines the cumulative displacement of the multi-segment image data corresponding to the inter-frame displacement based on the sub-pixel stitching function, and then performs bit division based on the coordinates corresponding to the cumulative displacement to determine the weight of neighboring pixels. Here, sub-pixel refers to pixels with a precision higher than one integer pixel unit, reaching the decimal level, such as 0.1-0.5 pixels. The sub-pixel stitching function is an image processing function that achieves sub-pixel-level accurate registration, misalignment compensation, and seamless fusion, used to correct minute inter-frame displacements and complete high-precision image stitching, including but not limited to Gaussian surface fitting, phase correlation methods, etc., which are not specifically limited in this embodiment. Then, the calculated inter-frame displacement sequence is applied to the sub-pixel stitching function, which transforms the inter-frame displacement sequence into absolute positions in the global coordinate system, obtaining the cumulative displacement of each frame image. Furthermore, based on the coordinates corresponding to this cumulative displacement, the number of bits is divided to determine the weight of neighboring pixels. That is, by decomposing each floating-point coordinate of the cumulative displacement into integer and fractional parts, the weight of the four surrounding integer neighboring pixels is calculated based on the fractional part. Then, the source pixel value is allocated and accumulated to these four neighbors according to this weight to obtain the weight of the neighboring pixels, thereby realizing sub-pixel precision bilinear resampling and accumulation of images.

[0046] It should be noted that after obtaining the weights of neighboring pixels, the current execution end performs normalization processing on the pixel image data based on bilinear interpolation weights and fusion weights. The total weights, including bilinear interpolation weights and preset spatial feathering fusion weights, are accumulated into a parallel weight map to obtain an initial stitched image. After stitching is completed, the weighted average normalization operation is completed by dividing the accumulated pixel image data by this initial stitched image to obtain stitched image data in the first format and stitched image data in the second format, that is, to generate a seamless and brightness-balanced 8-bit stitched preview image and a 16-bit stitched image.

[0047] In this embodiment, a variety of displacement estimation strategies and anti-jump constraints for track-oriented sub-pixel stitching methods can be adopted. Multiple inter-frame displacement estimation strategies such as physical displacement method and feature point matching method are provided. Track-oriented stitching is achieved by combining sub-pixel stitching function and edge feathering fusion. An anti-jump constraint mechanism for displacement is introduced to suppress drastic displacement jumps caused by instantaneous noise and solve the problem of insufficient accuracy of track-oriented stitching.

[0048] 104. Perform a first gradient fusion process on the three-channel image data in the stitched image data to obtain target color image data, and perform a second gradient fusion process on multiple stitched image data to obtain target wide-format image data.

[0049] In this embodiment, the current execution end employs a differentiated processing method for the stitched image data. First, it obtains three-channel image data from the stitched image data, namely an R(Red)G(Green)B(Blue) three-channel color image. Then, it performs a first gradient fusion process on this three-channel image data to obtain the target color image data, preferably a complete RGB color image without color borders. Simultaneously, the current execution end also performs a second gradient fusion process directly based on multiple stitched image data. Preferably, it uses the stitched image data of three long images for gradient fusion to obtain the final wide-format seamless stitched image.

[0050] In a specific image processing scenario, such as Figure 3 As shown, after reading the original raw / bin format file, performing segmented flat-field radiometric correction, and standardizing the data, multi-strategy inter-frame displacement estimation, displacement anti-jump constraint, and orbital sub-pixel stitching are performed based on 8-bit and 16-bit image data to output a high-quality long strip stitched image. Then, coarse matching of RGB three channels, fine matching of local affine transformation of strips, and gradient fusion are performed to output a colorless color image. At the same time, according to coarse and fine matching based on mutual information of adjacent long strips, radiometric difference correction, and linear gradient fusion, a wide-width seamless stitched image is output.

[0051] In another embodiment of this application, for further definition and explanation, the step of performing a first gradient fusion process based on the three-channel image data in the stitched image data to obtain the target color image data includes: Extract the three-channel image data of the stitched image data in the first format, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; Based on the preset strip height and preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated. Weighted fusion is performed based on the local transform field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data.

[0052] To achieve differentiated image processing objectives and meet diverse image processing needs, during the first gradient fusion processing based on the three-channel image data in the stitched image data, specifically, the current execution end first extracts the three-channel image data of the stitched image data in a first format, i.e., reads the long strip image data of the RGB three channels to ensure that the resolution and size of each channel image are consistent. Then, based on this three-channel image data, center region matching and alignment are performed to obtain coarsely matched image data. In a specific embodiment, center region matching and alignment refers to quickly estimating the global translation amount by performing matching in the center region of the image. The center region is a rectangular area extracted from the exact center of the three-channel image data according to a preset ratio, preferably 50%, to effectively utilize its typically high imaging quality and improve computational efficiency. The matching process can use AKAZE or SIFT algorithms to detect stable feature points such as corners and spots that are invariant to rotation and scaling within the center region. If it fails, it automatically reverts to using ORB features. Furthermore, k-nearest neighbor matching is performed on the feature points of two adjacent three-channel image data, and a ratio test is used to filter out high-confidence matching point pairs. The ratio of the ratio test can be set based on the filtering requirements, and this embodiment does not impose specific limitations. Finally, by calculating the median of the coordinate displacement of all high-quality matching point pairs, a sub-pixel-level global translation vector (dx, dy) with high robustness to noise and mismatches is obtained, which is the coarse-matched image data. If the feature point method fails due to texture loss or other reasons, the current execution end can also use the phase correlation method for translation estimation. This involves first performing preliminary correction on the images corresponding to the G and B channels, applying this translation vector to the G and B channel images, and then performing a sub-pixel-precision translation of the G and B channel images through affine transformation to align them with the reference R channel. Finally, all channels are uniformly cropped to retain a common effective field of view, ensuring that the output R, G, and B three-channel images are of consistent size and aligned in content, which is the coarse-matched image data.

[0053] After coarse matching is completed, the current execution end segments the coarsely matched image data along the cross-track direction based on a preset strip height and a preset overlap rate, obtaining multiple horizontal strip image data. It then calculates the local transformation field corresponding to the horizontal strip image data to achieve fine matching. In a specific embodiment, the coarsely matched image data is first segmented along the cross-track direction into multiple overlapping horizontal strips based on the preset strip height and preset overlap rate, resulting in multiple horizontal strip image data. These horizontal strip image data are then traversed, and for each pair of corresponding strips extracted from the channel to be corrected (e.g., G or B channel) and the reference channel (R channel), a local 2x3 affine transformation matrix is ​​independently calculated. This matrix is ​​calculated by performing feature point matching again within the strip and combining it with robust estimation algorithms such as RANSAC to accurately solve for the local transformation field, enabling simultaneous correction of translation, rotation, scaling, and shear. To ensure the continuity and stability of the algorithm, if registration fails due to texture loss or other reasons, the local affine matrix of the previous successful strip can be used. Finally, the set of local affine matrices calculated for all different strips constitutes a local transformation field covering the entire image. At this point, the multispectral band high-precision registration method based on strip-based local affine transformation employs a two-stage registration strategy from coarse to fine. It divides the image into overlapping horizontal strips and calculates independent local affine transformation matrices for each pair of strips. Through weighted feathering and fusion, the strips are recombined to achieve sub-pixel-level registration, fundamentally solving the color fringing problem.

[0054] Furthermore, when applying weighted fusion using a local transform field, each independent horizontal stripe image data is first geometrically corrected according to a specific affine matrix to obtain corrected image data. At this point, to eliminate seams between strips, a linearly gradient weight can be generated for the overlapping region of each strip, i.e., a linearly gradient weight, to accumulate the corrected image data and obtain the target color image data. Specifically, during weight accumulation, all independently corrected and weighted corrected image data are accumulated onto a global canvas and fused using weighted average normalization to generate a geometrically accurate and visually seamless target color image.

[0055] In a specific example of a gradient fusion process, after applying an independent affine transformation to each precisely matched horizontal stripe image data, linear gradient weights are generated based on the stripe index and a preset overlap height. At this point, the weight of the top overlapping region (excluding the first stripe) linearly increases from 0 to 1, the weight of the bottom overlapping region (excluding the last stripe) linearly decreases from 1 to 0, and the weight of the non-overlapping region remains at 1. The weights are then multiplied by the effective pixel mask after the affine transformation to obtain the linear gradient weights, thus masking invalid regions. Furthermore, when fusion is performed based on these linear gradient weights, weighted feathering fusion is preferred. This involves accumulating the pixel values ​​of each stripe after weight modulation to a global accumulation matrix, while simultaneously accumulating the weight values. Finally, normalization is achieved by dividing the global accumulation matrix by the total weight matrix, resulting in a seamless and smooth transition between adjacent stripes. The merging process refers to combining the registered and fused G and B single-channel grayscale images with the reference R channel grayscale image according to the color image channel synthesis rules to obtain a complete RGB color image without color borders, i.e., the target color image data.

[0056] In a specific example of implementing fine matching, the matching process does not perform a global single affine transformation. Instead, it uses a sliding strip window to estimate the affine transformation region by region. Seams are eliminated by fusion of gradually varying weights in overlapping areas, thus ensuring global alignment accuracy while allowing local images to adapt to non-rigid deformations. Specifically, first, three originally misaligned red, green, and blue monochrome images are roughly aligned, and black edges are cropped. Each image is then cut into narrow strips with overlaps. Instead of forcibly aligning the entire image at once, each strip undergoes fine alignment correction individually to ensure more accurate alignment in each local area. For overlapping areas of adjacent strips, a gradual transition from 0 to 1 transparency is used for a smooth transition, effectively erasing seams through feathering. This ensures overall alignment while adapting to minor local image deformations. Finally, the processed red, green, and blue channels are combined into a single image according to color image rules, resulting in a clear and complete color photograph without ghosting or color fringes—the target color image data.

[0057] In a specific example of estimating affine transformation, the embodiment of this application employs a method for calculating the strip-by-strip affine transformation matrix and a specific method for constructing the local transformation field. This method involves calculating the strip-by-strip affine transformation matrix for the independent strips after multi-channel image splitting, i.e., horizontal strip image data, using a feature-matching-based local parameter estimation method, and constructing a globally smooth local transformation field through region-by-region weighted interpolation. Specifically, firstly, SIFT feature points are extracted from the corresponding strip regions of the reference channel (R channel in the horizontal strip image data) and the channels to be registered (G and B channels in the horizontal strip image data), and bidirectional matching and mismatch removal are performed to obtain a high-precision set of corresponding feature point pairs. Based on the aforementioned feature point pairs, the 2D affine transformation matrix is ​​solved using the least squares method (LSM). The affine transformation model constructed independently for each band can be expressed as: ; in, For parameters of the combined transformation of rotation, scaling, and shearing, For the translation parameters, the optimal solution is obtained by minimizing the feature point projection error. To avoid splicing distortion caused by discontinuous transformations between strips, a globally continuous local transformation field can be constructed using a weighted linear interpolation method based on overlapping regions, with single-strip affine transformations as local bases. Specifically, using strip indices as positional weights, the affine transformation parameters of adjacent strips are smoothly interpolated, so that the transformation matrix of overlapping regions linearly transitions from the current strip to adjacent strips. Non-overlapping regions retain the independent affine transformation of the current strip, ultimately forming a globally smooth and locally accurate image transformation field, ensuring that subsequent strip fusion is without misalignment or abrupt changes.

[0058] In a specific gradient blending process, the gradient weight is obtained. Given a preset strip vertical height of H, a preset overlap height of M, and the current pixel row number of y, the strip weight W(y) is calculated using the following formula: .

[0059] Furthermore, for the top overlapping area (non-first strip), the middle non-overlapping area, and the bottom overlapping area (non-last strip), the gradient weight is multiplied by the effective pixel mask to mask invalid edge areas. Then, a weighted feathering fusion algorithm is used for strip stitching, where a weighted feathering fusion algorithm is set to... The pixel value of the k-th band after affine transformation. To correspond to the gradient weight mask, the global fused image F(x,y) satisfies the following condition: ; Here, ε is a minimum value used to avoid division by zero errors. Finally, through the above weighted accumulation and normalization, a seamless and smooth transition between adjacent stripes is achieved. In addition, during the three-channel merging, the R, G, and B channels each have a brightness level of 0-255. Color fusion refers to the precise matching and combination of the brightness values ​​of the same pixel position in the three channels to define the final color of each pixel. After combining all pixels in the entire image, a color image is generated.

[0060] In another embodiment of this application, for further definition and explanation, the step of performing a second gradient fusion process based on multiple stitched image data to obtain target wide-format image data includes: Extract multiple image data of preset length from the stitched image data of the second format, and perform coarse matching on the multiple image data of preset length based on the alignment method to obtain the corrected long strip image data; The overlapping regions of the elongated image data are fused to obtain reference image data; Radiometric correction is performed on the reference image data based on the regional pixel mean to obtain the target wide-area image data.

[0061] To achieve differentiated image processing objectives and meet diverse image processing needs, when performing a second gradient fusion process on multiple stitched image data to obtain target wide-format image data, specifically, multiple preset-length image data from the second-format stitched image data are extracted. This involves reading three long images output after edge feathering fusion, i.e., three preset-length image data, to ensure consistent image resolution and size. Furthermore, coarse matching is performed on the multiple preset-length image data based on an alignment method to obtain corrected long strip image data. At this point, a mutual information alignment estimation strategy is used to perform coarse matching in the data center region of the three preset-length image data. This involves aligning and estimating the narrow-band ROI regions above and below the stitching point of the three long images. The global translation offset (dx, dy) of the non-reference strip relative to the reference strip is obtained through mutual information maximization search, completing the initial correction of the non-reference strip and obtaining the long strip image data. Furthermore, each long strip image is horizontally divided into multiple overlapping strips. The local transformation relationship between the non-reference strip and the reference strip is calculated using a mutual information-based method. Simultaneously, for the segmented overlapping local strip regions, accurate local translation transformation parameters are obtained through local mutual information registration. Since this method is insensitive to linear radiation differences, it can meet the stitching requirements of sparse texture regions and obtain long strip image data.

[0062] In a specific example, the mutual information in this application embodiment serves as a similarity criterion between images, enabling automatic feature-free alignment and vertical stitching of multi-camera images. Specifically, the optimal translational displacement (dx, dy) between images is determined by searching for the maximum value of the mutual information, and then precise image stitching is completed based on this displacement. Specifically, firstly, the effective overlapping area between the reference image and the moving image to be stitched is extracted. The overlapping area is then uniformly converted into an 8-bit grayscale image. The mutual information value between the two grayscale images is calculated using a two-dimensional histogram. The mutual information calculation formula is expressed as: ; in, Let the gray levels of the two images be the joint probability density. , The mutual information values ​​represent the edge probability densities of grayscale in a single image. A higher mutual information value indicates a stronger statistical correlation and higher matching degree in the grayscale of the overlapping regions of the two images. Furthermore, a step-by-step search strategy combining coarse and fine search is employed. Within a preset translation range, all possible displacements (dx, dy) are traversed, and the mutual information value is calculated for each overlapping region under each displacement. The displacement that maximizes the mutual information value is selected as the optimal alignment displacement. Finally, the moving image is mapped to the coordinates according to this optimal translation displacement. Then, feathering fusion (e.g., using linear gradient weighted smooth transition) or cropping and stitching (e.g., removing overlapping regions to achieve hard-seam stitching) is applied to the overlapping regions of the images to stitch the moving image onto the canvas of the reference image, completing seamless and accurate stitching of multi-camera images and obtaining long strip image data. At this point, the robust cross-field multi-sensor image stitching method based on mutual information employs an alignment estimation strategy based on mutual information, combined with brightness histogram matching to correct radiometric differences, and achieves seamless stitching through linear gradient fusion of overlapping regions, solving the problem of cross-field stitching failure in sparse texture regions.

[0063] It should be noted that after obtaining the long strip image data, overlapping region fusion is performed on the long strip image data to obtain reference image data. At this point, the overlapping regions of adjacent images are first extracted. Spatial alignment is achieved by calculating the translational offset (dx, dy) between images based on mutual information maximization search. Bidirectional linear radiometric correction is then used to divide the long strip image data into a reference image and a registration image. The mean pixel value of the overlapping region between the reference image and the registration image is calculated. The median value of the two mean values ​​is then used as the target brightness. The gain coefficients of the two bands are calculated separately, and the global pixels are scaled to obtain the reference image data, thus eliminating radiometric differences. In this embodiment, after obtaining the reference image data, radiometric correction is performed on the reference image data based on the mean pixel value of the region to obtain the target wide-span image data. At this point, a method of constructing a vertical linear gradient weight based on the overlapping area can be adopted. Based on this weight, the pixels of the reference image and the image to be stitched are weighted and calculated to achieve a smooth transition in the overlapping area. Finally, a unified canvas is constructed based on the reference image, and all the images that have been aligned, corrected, and fused are pasted to the corresponding positions on the canvas one by one according to the calculated offset, completing the merging and stitching of all images and outputting target wide-format image data containing complete geographic information.

[0064] In a specific example of bidirectional linear radiometric correction, let the pixel mean of the overlapping region of the reference image be . The average pixel value of the overlapping region of the images to be registered is The median value of the two means is calculated as the target brightness. The calculation formula is expressed as: ; Specifically, the luminance gain coefficient of the reference image is calculated. , and the brightness gain coefficient of the image to be registered. , The gain coefficient is applied to the global pixels of the corresponding image to achieve brightness scaling and eliminate radiometric differences. When constructing a vertically linear gradient weight in the overlapping region for images that have undergone spatial alignment and radiometric correction, the vertical pixel height of the overlapping region can be set to H. Weight values ​​α∈[0,1] are generated sequentially from the reference image to the image to be registered along the vertical stitching direction, with the corresponding pixel weight in the reference image being 1. α, the pixel weight of the image to be registered is α, and the calculation formula can be expressed as: At this point, a weighted fusion calculation is performed on the pixels in the overlapping area of ​​the two images based on their weights. The calculation formula can be expressed as: ; in, The merged pixel values The baseline image pixel values, To achieve a smooth transition in overlapping areas, the pixel values ​​of the images to be registered are used, eliminating hard seams and ghosting. Finally, a unified pixel canvas is constructed based on the reference image. All images that have undergone spatial alignment, radiometric correction, and overlapping area fusion are successively mapped and pasted to the corresponding coordinate positions on the canvas according to the optimal translation offset (dx, dy) calculated above. This completes the integrated merging and stitching of all images, outputting a high-quality wide-format stitched image containing complete geographic information.

[0065] This application provides an image processing method. Compared with the prior art, this application provides an image processing method that acquires target image data to be processed; divides the target image data into multiple sub-segments according to the cross-track direction, and determines the inter-frame displacement of the standardized multi-segment image data according to different push-broom scenarios; fuses the multi-segment image data based on the inter-frame displacement to obtain stitched image data in different image formats; performs a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data; and performs a second gradient fusion process based on multiple stitched image data to obtain target wide-format image data. This achieves seamless stitching after cross-track stitching, improves stitching accuracy, ensures a color-edge-free image effect, and thus improves image processing accuracy.

[0066] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this application provides an image processing apparatus, such as... Figure 4 As shown, the device includes: Acquisition module 21 is used to acquire the target image data to be processed; The determination module 22 is used to divide the target image data into multiple sub-segments of image data according to the cross-track direction, and to determine the inter-frame displacement of the standardized multiple sub-segments of image data according to different push-broom scenarios. The fusion module 23 is used to fuse the multi-segment image data based on the inter-frame displacement to obtain stitched image data in different image formats; The processing module 24 is used to perform a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data, and to perform a second gradient fusion process based on multiple stitched image data to obtain target wide-area image data.

[0067] Furthermore, the device also includes: The denoising module is used to determine the flat field correction coefficient based on the stable column response curve of the target image data, and to denoise the multi-segment image data based on the flat field correction coefficient and the segmentation of the multi-segment image data. The mapping module is used to perform linear mapping on the denoised multi-segment image data according to a preset brightness range to obtain standardized multi-segment image data. The standardized multi-segment image data includes multi-segment image data in a first format and target image data in a second format.

[0068] Furthermore, The determining module calculates the inter-frame displacement of the standardized multi-segment image data based on ground resolution, camera frame rate, and aircraft speed when the push-broom scenario is a uniform push-broom scenario; when the push-broom scenario is a non-uniform push-broom scenario, it acquires feature points of the standardized multi-segment image data and determines the inter-frame displacement based on the feature points.

[0069] Furthermore, the determining module is also used to determine the inter-frame displacement corresponding to the previous frame as the inter-frame displacement of the current frame when the inter-frame displacement is greater than a preset threshold.

[0070] Furthermore, The fusion module is specifically used to determine the cumulative displacement of the multi-segment image data corresponding to the inter-frame displacement based on the sub-pixel stitching function, and to perform bit division based on the coordinates corresponding to the cumulative displacement to determine the weights of neighboring pixels; to perform re-sampling accumulation based on the weights of neighboring pixels and the coordinates to obtain pixel image data, and to perform normalization processing on the pixel image data based on bilinear interpolation weights and fusion weights to obtain stitched image data in a first format and stitched image data in a second format.

[0071] Further, the fusion module is specifically used to extract three-channel image data from the stitched image data of the first format, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; based on a preset strip height and a preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated; weighted fusion is performed based on the local transformation field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data.

[0072] Furthermore, the fusion module is specifically used to extract multiple image data of preset length from the stitched image data of the second format, and to perform coarse matching on the multiple image data of preset length based on the alignment method to obtain corrected long strip image data; to perform overlapping region fusion on the long strip image data to obtain reference image data; and to perform radiometric correction on the reference image data based on the regional pixel mean to obtain the target wide image data.

[0073] This application provides an image processing apparatus. Compared with the prior art, this application acquires target image data to be processed; divides the target image data into multiple sub-segments according to the cross-track direction, and determines the inter-frame displacement of the standardized multi-segment image data according to different push-broom scenarios; fuses the multi-segment image data based on the inter-frame displacement to obtain stitched image data of different image formats; performs a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data, and performs a second gradient fusion process based on multiple stitched image data to obtain target wide-width image data. This achieves the goal of seamless stitching after cross-track stitching, improves stitching accuracy, ensures a color-edge-free image effect, and thus improves image processing accuracy.

[0074] According to one embodiment of this application, a storage medium is provided, the storage medium storing at least one executable instruction that can perform the image processing method in any of the above method embodiments.

[0075] Figure 5 The diagram shows a structural schematic of a terminal according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the terminal.

[0076] like Figure 5 As shown, the terminal may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0077] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308.

[0078] Communication interface 304 is used to communicate with other network elements such as clients or other servers.

[0079] The processor 302 is used to execute program 310, specifically to perform the relevant steps in the above-described image processing method embodiments.

[0080] Specifically, program 310 may include program code that includes computer operation instructions.

[0081] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The terminal includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0082] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0083] Specifically, program 310 can be used to cause processor 302 to perform the following operations: Acquire the target image data to be processed; The target image data is divided into multiple sub-segments according to the cross-track direction, and the inter-frame displacement of the standardized multiple sub-segments is determined according to different push-broom scenarios. Based on the inter-frame displacement, the multi-segment image data is fused to obtain stitched image data in different image formats; A first gradient fusion process is performed on the three-channel image data in the stitched image data to obtain target color image data, and a second gradient fusion process is performed on multiple stitched image data to obtain target wide-format image data.

[0084] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0085] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An image processing method, characterized in that, include: Acquire the target image data to be processed; The target image data is divided into multiple sub-segments according to the cross-track direction, and the multiple sub-segments are standardized. The inter-frame displacement of the standardized multiple sub-segments is determined according to different push-broom scenarios. Based on the inter-frame shift, the multi-segment image data is fused to obtain multiple stitched image data including 8 bits and 16 bits. A first gradient fusion process is performed on the three-channel image data in the stitched image data to obtain target color image data, and a second gradient fusion process is performed on multiple stitched image data to obtain target wide-format image data. The process of fusing the multi-segment image data based on the inter-frame shift to obtain stitched image data in different image formats includes: The inter-frame displacement is determined based on the sub-pixel stitching function, which corresponds to the cumulative displacement of the multi-segment image data. Each floating-point coordinate of the cumulative displacement is decomposed into integer and fractional parts to determine the weight of neighboring pixels. Based on the weights of neighboring pixels and the coordinates, pixel image data is obtained by resampling and accumulating. Then, the pixel image data is normalized based on bilinear interpolation weights and fusion weights to obtain 8-bit multi-image stitched data and 16-bit multi-image stitched data. The first gradient fusion process based on the three-channel image data in the stitched image data to obtain the target color image data includes: Extract the three-channel image data from the 8-bit stitched image data, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; Based on the preset strip height and preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated. Weighted fusion is performed based on the local transform field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data; The second gradient fusion process based on multiple stitched image data to obtain the target wide-format image data includes: Extract multiple image data of preset lengths from the 16-bit stitched image data, and perform coarse matching on the multiple image data of preset lengths based on the alignment method to obtain the corrected long strip image data; The overlapping regions of the elongated image data are fused to obtain reference image data; Radiometric correction is performed on the reference image data based on the regional pixel mean to obtain the target wide-area image data.

2. The method according to claim 1, characterized in that, After dividing the target image data into multiple sub-segments according to the cross-track direction, the method further includes: The flat field correction coefficient is determined based on the stable column response curve of the target image data, and the multi-segment image data is denoised based on the flat field correction coefficient and the segmentation of the multi-segment image data. The denoised multi-segment image data is linearly mapped according to a preset brightness range to obtain standardized multi-segment image data, which includes 8-bit multi-segment image data and 16-bit multi-segment image data.

3. The method according to claim 2, characterized in that, The process of determining the inter-frame shift of the standardized multi-segment image data according to different pushbroom scenarios includes: When the push-broom scenario is a uniform push-broom scenario, the inter-frame displacement of the standardized multi-segment image data is calculated based on the ground resolution, camera frame rate and aircraft speed. When the push-broom scenario is a non-uniform push-broom scenario, the feature points of the standardized multi-segment image data are obtained, and the inter-frame displacement is determined based on the feature points.

4. The method according to claim 3, characterized in that, The method further includes: When the inter-frame displacement is greater than a preset threshold, the inter-frame displacement corresponding to the multi-segment image of the previous frame image data is determined as the inter-frame displacement corresponding to the multi-segment image of the target image data to be processed in the current frame.

5. An image processing apparatus, characterized in that, include: The acquisition module is used to acquire the target image data to be processed; The determination module is used to divide the target image data into multiple sub-segments of image data according to the cross-track direction, and to standardize the multiple sub-segments of image data, and to determine the inter-frame displacement of the standardized multiple sub-segments of image data according to different push-broom scenarios. The fusion module is used to fuse the multi-segment image data based on the inter-frame displacement to obtain multiple stitched image data including 8 bits and 16 bits; The processing module is used to perform a first gradient fusion process based on the three-channel image data in the stitched image data to obtain target color image data, and to perform a second gradient fusion process based on multiple stitched image data to obtain target wide-area image data. The fusion module is specifically used to determine the cumulative displacement of the inter-frame displacement corresponding to the multi-segment image data based on the sub-pixel stitching function, and to decompose each floating-point coordinate of the cumulative displacement into integer and fractional parts to determine the weight of neighboring pixels; to resample and accumulate the coordinates based on the weight of neighboring pixels to obtain pixel image data, and to normalize the pixel image data based on bilinear interpolation weights and fusion weights to obtain 8-bit multi-segment image data and 16-bit multi-segment image data; The processing module is specifically used to extract the three-channel image data of the 8-bit stitched image data, and perform center region matching and alignment based on the three-channel image data to obtain coarse matching image data; Based on the preset strip height and preset overlap rate, the coarse matching image data is segmented according to the cross-track direction to obtain multiple horizontal strip image data, and the local transformation field corresponding to the horizontal strip image data is calculated; weighted fusion is performed based on the local transformation field to obtain corrected image data, and the corrected image data is weighted and accumulated based on linear gradient weights to obtain target color image data; The processing module is further configured to extract multiple image data of preset lengths from the 16-bit stitched image data, and perform coarse matching on the multiple image data of preset lengths based on the alignment method to obtain the corrected strip image data. The overlapping regions of the long strip image data are fused to obtain reference image data; radiometric correction is performed on the reference image data based on the mean pixel value of the region to obtain the target wide-span image data.

6. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.

Images