On-line correction method and system for surface regression of thermal radiation effect of hypersonic vehicle

Abstract

The present invention discloses a method and system for online correction of thermal radiation effect surface regression of hypersonic aircraft, wherein the method comprises the steps of: obtaining a degraded image containing thermal radiation effect; obtaining a dark channel image of the degraded image, and performing semantic segmentation; filtering and calculating a preliminary estimated thermal radiation effect image of the semantic block; performing quadtree block division, and constructing a binary third-order polynomial surface regression model; fitting image blocks, splicing semantic blocks, and then splicing all semantic blocks to obtain a complete regression fitting surface as a thermal radiation effect bias field, and obtaining a potential clear image; through iterative regression fitting surface and potential clear image, until the surface regression model converges or reaches the iteration termination condition, the potential clear image obtained after the last iteration is the final clear image after correction. The present invention can complete the task of removing thermal radiation effect with high quality, and greatly improve the image signal-to-noise ratio.

Description

Technical Field

[0001] The present invention relates to the field of image aerodynamic optical thermal radiation effect correction, and in particular to a surface regression online correction method and system for thermal radiation effect of a hypersonic aircraft. Background Art

[0002] When an aircraft flies at hypersonic speed in a dense atmosphere, the detector optical hood will be affected by the airflow. In the boundary layer near the hood surface, the kinetic energy of the airflow is converted into thermal energy, causing the optical window temperature to rise rapidly. This phenomenon is called the aerodynamic optical thermal radiation effect. This effect seriously affects the detection efficiency of the imaging detection system. It forms radiation interference on the imager, increases the image background brightness, and significantly reduces the signal-to-noise ratio of the captured image, thereby hindering subsequent target detection and identification.

[0003] At present, there are some effective methods to correct the aero-optical thermal radiation effect in the image, such as replacing the infrared window material with high transmission and low emissivity, applying thermal protection coating on the surface of optical devices, using window cooling technology, etc. Although the above technologies can effectively prevent the saturation of optical equipment, these physical implementation methods are relatively complicated to operate and expensive, and there will still be residual thermal radiation effects to varying degrees.

[0004] Currently, most of the thermal radiation effect correction methods available at the algorithm level are spline surface fitting correction. The thermal radiation effects they target are mostly generated by a single heat source and have overall consistency in the degraded image. However, when the heat source is not unique and its distribution is more scattered, there will be problems such as more complex models, lower fitting accuracy, and poor local adaptability. Summary of the invention

[0005] The main purpose of the present invention is to provide a method and system for online correction of thermal radiation effect surface regression of hypersonic aircraft, which can complete the task of removing thermal radiation effect with high quality and greatly improve the image signal-to-noise ratio.

[0006] The technical solution adopted by the present invention is:

[0007] A method for online correction of thermal radiation effect surface of a hypersonic aircraft by regression is provided, which is characterized by comprising the following steps:

[0008] S1, obtaining a degraded image containing thermal radiation effects;

[0009] S2, obtaining a dark channel image of the degraded image, and performing semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations;

[0010] S3, filtering the multiple semantic blocks, and calculating a preliminary estimated thermal radiation effect image of each semantic block;

[0011] S4, based on regional uniformity, the preliminary estimated thermal radiation effect image of each semantic block is divided into quadtree blocks, and a binary third-order polynomial surface regression model is constructed for each image block after quadtree block division;

[0012] S5, by adding the constraints that G0 and G1 are continuous between adjacent image blocks, the polynomial coefficients of each image block are solved by the least square method, each image block is fitted, the image blocks are spliced ​​to fit the corresponding semantic blocks, and then all the semantic blocks are spliced ​​to obtain a complete regression fitting surface as a thermal radiation effect bias field;

[0013] S6, subtracting the degraded image containing the thermal radiation effect from the thermal radiation effect bias field and performing grayscale stretching to obtain a potential clear image;

[0014] S7. Fitting the surface and the potential clear image through iterative regression until the surface regression model converges or the iteration termination condition is reached, and the potential clear image obtained after the last iteration is the final clear image after correction.

[0015] Following the above technical solution, in step S3, a reference-free gradient operator is specifically used to estimate the severity of thermal radiation of each semantic block, and adaptive filtering is performed on semantic blocks of different degrees.

[0016] Following the above technical solution, step S3 specifically includes: calculating the gradient strength and gradient direction of each semantic block; calculating the average gradient strength and gradient variance of each semantic block; using the linear combination formula to estimate the severity of thermal radiation of each semantic block, and performing adaptive L0smooth filtering to calculate the preliminary estimated thermal radiation effect image of each semantic block.

[0017] Following the above technical solution, in step S7, the surface and the potential clear image are fitted by progressive iterative regression.

[0018] Following the above technical solution, step S2 specifically includes: extracting each pixel coordinate and corresponding pixel value of the degraded image, combining them into a feature vector, normalizing the feature vector, and performing semantic segmentation according to the value of the feature vector after the normalization.

[0019] Following the above technical solution, the quadtree partitioning in step S4 is specifically as follows: quadtree partitioning is performed on a semantic block to obtain four block areas, a threshold is set, and the standard deviation of the pixel values ​​of one of the block areas is calculated. If the standard deviation is greater than the threshold, it is considered that the block area is not uniform enough, and the partitioning is continued until the block is divided to the minimum size.

[0020] The present invention also provides a hypersonic aircraft thermal radiation effect curved surface regression online correction system, comprising:

[0021] An image acquisition module, used for acquiring a degraded image containing a thermal radiation effect;

[0022] A semantic segmentation module is used to obtain a dark channel image of the degraded image and perform semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations;

[0023] A filtering module, used for filtering the multiple semantic blocks and calculating a preliminary estimated thermal radiation effect image of each semantic block;

[0024] A quadtree block module is used to perform quadtree block division on the preliminary estimated thermal radiation effect image of each semantic block based on regional uniformity, and to construct a binary third-order polynomial surface regression model for each image block after quadtree block division;

[0025] The fitting module is used to fit each image block by adding the constraints that G0 and G1 are continuous between adjacent image blocks, solving the polynomial coefficients of each image block using the least squares method, splicing the image blocks into corresponding semantic blocks, and then splicing all the semantic blocks to obtain a complete regression fitting surface as a thermal radiation effect bias field;

[0026] The correction module is used to subtract the degraded image containing thermal radiation effect from the thermal radiation effect bias field and perform grayscale stretching to obtain a potential clear image; the surface and the potential clear image are fitted through iterative regression until the surface regression model converges or the iteration termination condition is reached. The potential clear image obtained after the last iteration is the final clear image after correction.

[0027] Following the above technical solution, the correction module specifically fits the surface and the potential clear image through progressive iterative regression.

[0028] Following the above technical solution, when the quadtree blocking module performs quadtree blocking, it specifically performs quadtree blocking on a semantic block to obtain four block areas, sets a threshold, and calculates the standard deviation of the pixel values ​​of one of the block areas. If the standard deviation is greater than the threshold, it is considered that the block area is not uniform enough, and the blocking continues until the block is divided to the minimum size.

[0029] The present invention also provides a computer storage medium, which stores a computer program that can be executed by a processor, and the computer program executes the hypersonic aircraft thermal radiation effect surface regression online correction method described in the above technical solution.

[0030] The beneficial effects of the present invention are as follows: the present invention targets radiation interference under multiple heat sources, converts the degraded image into a dark channel image and then performs semantic segmentation, segments the areas under the influence of different heat sources, and then performs quadtree block division on the preliminary estimated thermal radiation effect image of each semantic block based on regional uniformity, and performs constrained polynomial regression fitting on each image block after quadtree block division, and then fits the spliced ​​image blocks into corresponding semantic blocks, and finally splices all semantic blocks to obtain a complete regression fitting surface, thereby obtaining a thermal radiation effect bias field. The present invention can flexibly fit complex image features, perform corrections for the characteristics of different regions, and avoid information loss or over-smoothing problems caused by global processing; and the present invention is simple to implement and has generalization, can complete the task of removing thermal radiation effects with high quality, greatly improve the image signal-to-noise ratio, and the residual thermal radiation effect of the corrected image is no longer obvious, solving the problem that the existing aerodynamic optical effect correction method is either complex to implement or still has obvious residual thermal radiation effects.

[0031] Furthermore, the iterative method introduces a progressive iteration framework and adopts a step-by-step iteration approach. Each step only needs to solve a relatively simple least squares problem. Compared with directly solving complex global optimization problems, it has higher computational efficiency and greater flexibility.

[0032] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 It is a flow chart of a method for online correction of thermal radiation effect curved surface regression of a hypersonic aircraft according to an embodiment of the present invention;

[0035] Figure 2 (a) is a degradation image containing different degrees of thermal radiation effects, Figure 2 (b) is a dark channel image obtained by processing the degraded image, and Figure 2 (c) is the result of semantic segmentation of the thermal radiation degraded image;

[0036] Figure 3 (a)-Figure 3 (d) are schematic diagrams of the preliminary estimated thermal radiation effect images of each semantic image block after filtering;

[0037] FIG4(a) to FIG4(d) are schematic diagrams of performing quadtree partitioning on each semantic image block;

[0038] Figure 5(a) is the thermal radiation bias field of the entire image after polynomial regression fitting and splicing, Figure 5(b) is the final correction image after 20 iterations of the algorithm, and Figure 5(c) is the actual clear image. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] It should be noted that the illustrations provided in the embodiments of the present invention are only used to illustrate the basic concept of the present invention in a schematic manner. Therefore, the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout type may also be more complicated.

[0041] In the present invention, it is also necessary to explain that, if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, if the terms "first" and "second" appear, they are only used for description and distinction purposes, and cannot be understood as indicating or implying relative importance.

[0042] Example 1

[0043] like Figure 1 As shown, the hypersonic aircraft thermal radiation effect surface regression online correction method of this embodiment includes the following steps:

[0044] S1, obtaining a degraded image containing thermal radiation effects;

[0045] S2, obtaining a dark channel image of the degraded image, and performing semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations;

[0046] S3, filtering the multiple semantic blocks, and calculating a preliminary estimated thermal radiation effect image of each semantic block;

[0047] S4, based on regional uniformity, the preliminary estimated thermal radiation effect image of each semantic block is divided into quadtree blocks, and a binary third-order polynomial surface regression model is constructed for each image block after quadtree block division;

[0048] S5, by adding the constraints that G0 and G1 are continuous between adjacent image blocks, the polynomial coefficients of each image block are solved by the least square method, each image block is fitted, the image blocks are spliced ​​to fit the corresponding semantic blocks, and then all the semantic blocks are spliced ​​to obtain a complete regression fitting surface as a thermal radiation effect bias field;

[0049] S6, subtracting the degraded image containing the thermal radiation effect from the thermal radiation effect bias field and performing grayscale stretching to obtain a potential clear image;

[0050] S7. Fit the surface and the potential clear image through progressive iterative regression until the surface regression model converges or the iteration termination condition is reached, and the potential clear image obtained after the last iteration is the final clear image after correction.

[0051] Furthermore, step S2 specifically includes: extracting each pixel coordinate and corresponding pixel value of the degraded image, combining them into a feature vector, normalizing the feature vector, and performing semantic segmentation according to the value of the feature vector after the normalization.

[0052] Furthermore, in step S3, a reference-free gradient operator is specifically used to estimate the severity of thermal radiation of each semantic block, and adaptive filtering is performed on semantic blocks of different degrees.

[0053] This step S3 specifically includes: calculating the gradient strength and gradient direction of each semantic block; calculating the average gradient strength and gradient variance of each semantic block; using a linear combination formula to estimate the severity of thermal radiation of each semantic block, and performing adaptive L0smooth filtering to calculate a preliminary estimated thermal radiation effect image of each semantic block.

[0054] Among them, the quadtree blocking in step S4 is specifically as follows: quadtree blocking is performed on a semantic block to obtain four block areas, a threshold is set, and the standard deviation of the pixel values ​​of one of the block areas is calculated. If the standard deviation is greater than the threshold, it is considered that the block area is not uniform enough, and the blocking is continued until the block is divided to the minimum size.

[0055] Furthermore, in step S7, the curved surface and the potential clear image are fitted by progressive iterative regression.

[0056] This embodiment mainly performs semantic block processing on the image according to different heat sources, and performs quadtree block processing on the preliminary estimated thermal radiation effect image of different semantic blocks, and performs constrained polynomial regression fitting on each image block after quadtree block processing. It can flexibly fit complex image features, perform corrections according to the characteristics of different regions, and avoid information loss or over-smoothing problems caused by global processing.

[0057] Example 2

[0058] This embodiment is based on Embodiment 1, and a more optimal implementation scheme is selected for each step on the basis of Embodiment 1.

[0059] Specifically, the method in S2 is:

[0060] The degraded image z with thermal radiation effect is obtained, as shown in Figure 2 (a). The dark channel image is obtained for the degraded image z with thermal radiation effect. , as shown in Figure 2 (b), the principle is:

[0061]

[0062] in, So Centered at The local window of r The size of is set to 15, is the image z in the local window Inner pixel The gray value of .

[0063] For dark channel images Semantic segmentation is performed. According to the brightness distribution of semantic segmentation, the non-uniform semantic thermal radiation degraded image blocks affected by different degrees of thermal radiation in the degradation image are separated. Specifically: Extract Each pixel coordinate and corresponding pixel value in the image are combined into a feature vector, and the feature vector is standardized:

[0064]

[0065] in is the eigenvector i The original value of is the mean, is the standard deviation, is the standardized value, and then non-uniformly divided according to the following formula:

[0066]

[0067] in is the number of blocks, is assigned to j Block i feature vectors, It is j The center of the block.

[0068] As shown in FIG2 (c), the non-uniform semantic thermal radiation degraded image blocks affected by different degrees of thermal radiation are separated. In this embodiment, a total of 5 image blocks are separated, of which 4 are affected by different degrees of thermal radiation, and the remaining 1 is regarded as not affected by thermal radiation. However, in order to splice a complete thermal radiation bias field, it is only filtered subsequently.

[0069] The method in step S3 is specifically:

[0070] For each semantic block, i.e., the semantic thermal radiation degraded image block, the severity of thermal radiation is estimated using the reference-free gradient operator. Specifically, the gradient strength of each semantic thermal radiation degraded image block is calculated. and the gradient direction :

[0071]

[0072]

[0073] in and Represent the horizontal gradient and vertical gradient respectively.

[0074] Calculate the average gradient strength of each semantic thermal radiation degraded image patch and the gradient variance :

[0075]

[0076]

[0077] in n Represents the number of pixels in the image block.

[0078] Using the linear combination formula Estimate the severity of the heat radiation of this part. and They are set to 0.65 and 0.35 respectively and adaptive L0smooth filtering is performed. Then, the preliminary estimated thermal radiation effect image of each semantic image block after filtering is calculated, as shown in Figure 3 (a) to Figure 3 (d).

[0079] The method in step S4 is specifically as follows:

[0080] The thermal radiation effect image of each semantic image block after filtering is preliminarily estimated. The image is divided into quadtree blocks based on regional uniformity and the threshold is set. , if set to 20, calculate the standard deviation of the pixel values ​​in the area , the specific calculation formula is:

[0081]

[0082] in is the pixel value in the region, is the mean value of pixels in the region.

[0083] If the standard deviation Greater than the threshold , the area is considered not uniform enough, and the block division is continued until the minimum size is reached to improve the accuracy of subsequent regression fitting. The minimum size in this example is 30×30. The block division results are shown in Figure 4 (a) to Figure 4 (d). A binary third-order polynomial surface regression model is established:

[0084]

[0085] in, are the coefficients of the polynomial, i and j Respectively x and y The number of times, and, satisfying i+j ≦3.

[0086] Construction Design Matrix A :

[0087]

[0088] Add constraints G0 and G1 between adjacent image blocks, where G0 continuity means that the function values ​​are equal on the boundaries of adjacent image blocks. G1 continuity means that the first-order partial derivatives of the function are equal on the boundaries of adjacent image blocks.

[0089] Combine the least squares method and constraints to solve the surface expression coefficient matrix of each image block , whose least squares problem is formulated as minimizing the following objective function:

[0090]

[0091] in z represents the degraded image, Indicates a clear image.

[0092] Its least squares solution is:

[0093]

[0094] Substituting the solved coefficients into the surface regression model can obtain the regression fitting expression of the image block.

[0095] The method in step S5 is specifically as follows:

[0096] The expression obtained in step S4 is the regression fitting surface of the corresponding image block after quadtree partitioning. These surfaces are spliced ​​to obtain the regression fitting surface of each semantic image block as the thermal radiation effect bias field of the semantic image block. , and then splicing the regression fitting surface of the semantic image block to obtain the thermal radiation effect bias field of the entire image , as shown in Figure 5(a).

[0097] The method in step S6 is specifically as follows:

[0098] In order to ensure that the model has a suitable convergence speed in the subsequent iteration process, the Weighted, the thermal radiation degradation map z and the weighted Subtract to get the initial correction image ,Right now:

[0099]

[0100] in is the weight, within the range The larger the value, the faster the model converges, but it may remove effective information in the image that does not belong to the thermal radiation effect, affecting subsequent iterations. The smaller the value of , the better the image quality can be guaranteed during the iteration process, but it will affect the convergence speed of the model.

[0101] right Perform grayscale stretching to obtain a potentially clear image ,Right now:

[0102]

[0103] in Represents images The maximum and minimum values ​​of the pixels in .

[0104] The method in step S7 is specifically as follows:

[0105] Take the progressive iteration method and take the potential clear image obtained by S6 in this iteration As the degraded image input for the next iteration, steps S2 to S6 are repeated until the algorithm converges or the iteration termination condition is reached. The potential clear image obtained after the last iteration is output as the final corrected image. As shown in Figure 5 (b), it is the final corrected image after 20 iterations, and as shown in Figure 5 (c), it is the actual clear image. It can be seen that the correction result is of high quality.

[0106] Example 3

[0107] In order to implement the above method embodiment, the present invention also provides a hypersonic aircraft thermal radiation effect surface regression online correction system, comprising:

[0108] An image acquisition module, used for acquiring a degraded image containing a thermal radiation effect;

[0109] A semantic segmentation module is used to obtain a dark channel image of the degraded image and perform semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations;

[0110] A filtering module, used for filtering the multiple semantic blocks and calculating a preliminary estimated thermal radiation effect image of each semantic block;

[0111] A quadtree block module is used to perform quadtree block division on the preliminary estimated thermal radiation effect image of each semantic block based on regional uniformity, and to construct a binary third-order polynomial surface regression model for each image block after quadtree block division;

[0112] The fitting module is used to fit each image block by adding the constraints that G0 and G1 are continuous between adjacent image blocks, solving the polynomial coefficients of each image block using the least squares method, splicing the image blocks into corresponding semantic blocks, and then splicing all the semantic blocks to obtain a complete regression fitting surface as a thermal radiation effect bias field;

[0113] The correction module is used to subtract the degraded image containing thermal radiation effect from the thermal radiation effect bias field and perform grayscale stretching to obtain a potential clear image; the surface and the potential clear image are fitted through iterative regression until the surface regression model converges or the iteration termination condition is reached. The potential clear image obtained after the last iteration is the final clear image after correction.

[0114] Furthermore, the correction module specifically fits the surface and the potential clear image through progressive iterative regression.

[0115] Furthermore, when the quadtree blocking module performs quadtree blocking, it specifically performs quadtree blocking on a semantic block to obtain four block areas, sets a threshold, and calculates the standard deviation of the pixel values ​​of one of the block areas. If the standard deviation is greater than the threshold, it is considered that the block area is not uniform enough, and the blocking continues until the block is blocked to the minimum size.

[0116] Each module is mainly used to implement each step of the above method embodiment, which will not be described one by one here.

[0117] The present application also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App application store, etc., on which a computer program is stored, and when the program is executed by a processor, a corresponding function is realized. When the computer-readable storage medium of this embodiment is executed by a processor, the method embodiment of the method is implemented. The method of online correction of the surface regression of the thermal radiation effect of the hypersonic aircraft.

[0118] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0119] The order of execution of each step in the above embodiment does not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0120] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for online correction of thermal radiation effect surface regression of hypersonic aircraft, characterized in that: The following steps are involved: S1, obtaining a degraded image containing thermal radiation effects; S2, obtaining a dark channel image of the degraded image, and performing semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations; S3, filtering the multiple semantic blocks, and calculating a preliminary estimated thermal radiation effect image of each semantic block; S4, based on regional uniformity, the preliminary estimated thermal radiation effect image of each semantic block is divided into quadtree blocks, and a binary third-order polynomial surface regression model is constructed for each image block after quadtree block division; S5, by adding the constraints that G0 and G1 are continuous between adjacent image blocks, the polynomial coefficients of each image block are solved by the least square method, each image block is fitted, the image blocks are spliced ​​to fit the corresponding semantic blocks, and then all the semantic blocks are spliced ​​to obtain a complete regression fitting surface as a thermal radiation effect bias field; S6, subtracting the degraded image containing the thermal radiation effect from the thermal radiation effect bias field and performing grayscale stretching to obtain a potential clear image; S7. Fitting the surface and the potential clear image through iterative regression until the surface regression model converges or the iteration termination condition is reached, and the potential clear image obtained after the last iteration is the final clear image after correction.

2. The method for online correction of thermal radiation effect surface regression of hypersonic aircraft according to claim 1, characterized in that: In step S3, a reference-free gradient operator is specifically used to estimate the severity of thermal radiation of each semantic block, and adaptive filtering is performed on semantic blocks of different degrees.

3. The method for online correction of thermal radiation effect surface regression of hypersonic aircraft according to claim 1, characterized in that: Step S3 specifically includes: calculating the gradient strength and gradient direction of each semantic block; calculating the average gradient strength and gradient variance of each semantic block; estimating the severity of thermal radiation of each semantic block using a linear combination formula, and performing adaptive L0smooth filtering to calculate a preliminary estimated thermal radiation effect image of each semantic block.

4. The method for online correction of thermal radiation effect surface regression of hypersonic aircraft according to claim 1, characterized in that: In step S7, the curved surface and the potential clear image are fitted by progressive iterative regression.

5. The method for online correction of thermal radiation effect surface regression of hypersonic aircraft according to claim 1, characterized in that: Step S2 specifically includes: extracting each pixel coordinate and corresponding pixel value of the degraded image, combining them into a feature vector, normalizing the feature vector, and performing semantic segmentation according to the value of the feature vector after the normalization.

6. The method for online correction of thermal radiation effect surface regression of hypersonic aircraft according to claim 1, characterized in that: The quadtree partitioning in step S4 is specifically as follows: a semantic block is partitioned by a quadtree to obtain four partition areas, a threshold is set, and the standard deviation of the pixel values ​​in one of the partition areas is calculated. If the standard deviation is greater than the threshold, it is considered that the partition area is not uniform enough, and the partitioning is continued until the partition reaches the minimum size.

7. A hypersonic aircraft thermal radiation effect surface regression online correction system, characterized in that: include: An image acquisition module, used for acquiring a degraded image containing a thermal radiation effect; A semantic segmentation module is used to obtain a dark channel image of the degraded image and perform semantic segmentation on the dark channel image to obtain multiple semantic blocks for multiple heat source segmentations; A filtering module, used for filtering the multiple semantic blocks and calculating a preliminary estimated thermal radiation effect image of each semantic block; A quadtree block module is used to perform quadtree block division on the preliminary estimated thermal radiation effect image of each semantic block based on regional uniformity, and to construct a binary third-order polynomial surface regression model for each image block after quadtree block division; The fitting module is used to fit each image block by adding the constraints that G0 and G1 are continuous between adjacent image blocks, solving the polynomial coefficients of each image block using the least squares method, splicing the image blocks into corresponding semantic blocks, and then splicing all the semantic blocks to obtain a complete regression fitting surface as a thermal radiation effect bias field; The correction module is used to subtract the degraded image containing thermal radiation effect from the thermal radiation effect bias field and perform grayscale stretching to obtain a potential clear image; the surface and the potential clear image are fitted through iterative regression until the surface regression model converges or the iteration termination condition is reached. The potential clear image obtained after the last iteration is the final clear image after correction.

8. The hypersonic vehicle thermal radiation effect surface regression online correction system according to claim 7, characterized in that: The correction module specifically fits the surface and the potential clear image through progressive iterative regression.

9. The hypersonic vehicle thermal radiation effect surface regression online correction system according to claim 7, characterized in that: When the quadtree block module performs quadtree block, it specifically performs quadtree block on a semantic block to obtain four block areas, sets a threshold, and calculates the standard deviation of the pixel values ​​of one of the block areas. If the standard deviation is greater than the threshold, it is considered that the block area is not uniform enough, and the block continues until the block is divided to the minimum size.

10. A computer storage medium, characterized in that: A computer program executable by a processor is stored therein, and the computer program executes the hypersonic aircraft thermal radiation effect curved surface regression online correction method according to any one of claims 1 to 6.

Images