Method and system for processing underwater environment image of nuclear power plant

Abstract

The application discloses a nuclear power plant underwater environment image processing method and system. The method comprises the following steps: converting a video stream of a nuclear power plant underwater environment into an image frame sequence, and pre-processing each image frame; determining the translation component of two adjacent image frames according to the pre-processed image frame sequence, and aligning each image frame to a reference coordinate system; determining the contrast weight, saturation weight and exposure weight of each aligned image frame, and determining the weighted image of the image frame; and performing fusion processing on each weighted image to obtain a panoramic splicing image. The technical scheme of the application can realize robust suppression of various random disturbances and robust recovery of moving targets, highlight real details and suppress artifacts, so that a clear, stable and measurable underwater panoramic image is obtained, and the efficiency and reliability of underwater measurement tasks are improved.

Description

Technical Field

[0001] This application relates to the field of nuclear power, and in particular to a method and system for processing underwater environmental images of nuclear power plants. Background Technology

[0002] In critical operations at nuclear power plants, such as fuel assembly loading and unloading, reactor component inspection, pool maintenance, and decommissioning, visual inspection and image measurement are indispensable key methods. However, unlike ordinary industrial environments, nuclear power plant fuel pools and related underwater areas are constantly subjected to multi-physics coupling environments, including: 1. Underwater fluid disturbance field: Under the influence of mechanical motion and heat flow, water forms microscale eddies or local flows, causing instantaneous deformation, local stretching, jitter, or blurring of the image; 2. Underwater scattering and absorption: The turbidity of the water itself leads to decreased image contrast, obscured details, and softened edges; 3. Radiation interference field: The interaction between high-energy particles and imaging devices can cause random flicker noise, bright spots, streaks, or local exposure anomalies.

[0003] The superposition of these various physical disturbances, characterized by high randomness, high frequency, and spatial non-uniformity, significantly degrades image quality. This forces critical underwater measurement tasks to still rely on manual compensation or repeated imaging, impacting efficiency and reliability. Therefore, a novel solution is urgently needed that can maintain stable, measurable, and disturbance-resistant imaging even in complex multiphysics coupling environments. Summary of the Invention

[0004] The technical problem to be solved by this application is to provide a method and system for processing underwater environmental images of nuclear power plants, addressing the aforementioned technical deficiencies in the existing technology.

[0005] The technical solution adopted by this application to solve its technical problem is: a method for processing underwater environment images of nuclear power plants, including: Step S10: Convert the acquired video stream of the underwater environment of the nuclear power plant into a sequence of image frames, and preprocess each image frame. Step S20: Based on the preprocessed image frame sequence, determine the translation components of two adjacent image frames, and align each image frame to the reference coordinate system based on the translation components of two adjacent image frames. Step S30: For each aligned image frame, determine the pixel-level contrast weight, saturation weight, and exposure weight of the image frame, and determine the weighted image of the image frame based on the contrast weight, saturation weight, and exposure weight. Step S40: Perform fusion processing on each weighted image to obtain a panoramic stitched image.

[0006] Optionally, in step S10, the preprocessing of each image frame includes: For each image frame, perform color correction and / or outlier removal.

[0007] Optionally, in step S20, determining the translation components of two adjacent image frames includes: Step S21: Perform frequency domain phase matching on the grayscale images of two adjacent preprocessed image frames to obtain the cross-correlation surface of the two adjacent image frames. Step S22: Determine the peak value of the cross-correlation surface, and calculate the translation component and translation confidence level based on the peak value; Step S23: For image frames whose translation confidence is less than the threshold, discard the calculated translation component and estimate its translation component relative to the previous image frame using cubic spline interpolation.

[0008] Optionally, in step S21, for each preprocessed image frame, the following is performed: Step S211: Use a two-dimensional window function to perform pixel-level weighting on the grayscale image of the image frame, and perform Fourier transform on the weighted grayscale image to obtain the frequency domain spectrum of the image frame. Step S212: Calculate the normalized cross power spectrum based on the frequency domain spectrum of the image frame and the frequency domain spectrum of the previous image frame, and perform an inverse Fourier transform on the normalized cross power spectrum to obtain the cross-correlation surface between the image frame and the previous image frame.

[0009] Optionally, in step S212, the normalized cross power spectrum is calculated according to formula 2-1; Formula 2-1 in, Normalized cross-power spectrum; Let be the frequency domain spectrum of the t-th image frame; Let be the frequency domain spectrum of the (t-1)th image frame; It is a constant.

[0010] Optionally, in step S20, aligning each image frame to the reference coordinate system based on the translation components of two adjacent image frames includes: Step S24: For each image frame, the translation components of the image frame and each image frame preceding it relative to their respective previous image frame are accumulated to obtain the total translation of the image frame relative to the reference frame. Step S25: For each image frame, construct a translation matrix according to its corresponding total translation amount, and translate the color image of the image frame according to the translation matrix to align the image frame to the reference coordinate system.

[0011] Optionally, in step S30, determining the contrast weight, saturation weight, and exposure weight at the pixel level of the image frame includes: Step S31: Calculate the pixel-level contrast weights of the image frame according to Formula 3-1; Formula 3-1 in, The coordinates of the t-th image frame after alignment are ( The contrast weight of the pixels; This refers to the t-th image frame after alignment. This is a grayscale conversion function; For the Laplace operator; Step S32: Calculate the pixel-level saturation weight of the image frame according to Formula 3-2; Formula 3-2 in, The coordinates of the t-th image frame after alignment are ( The saturation weight of the pixels; The coordinates of the t-th image frame after alignment are ( The pixel value of the pixel in channel c. The coordinates of the t-th image frame are ( The average pixel value of a pixel across the three channels; Step S33: Calculate the pixel-level exposure weight of the image frame according to formula 3-3; Formula 3-3 in, The coordinates of the t-th image frame after alignment are ( Exposure weights of pixels; It is a Gaussian function; The coordinates of the t-th image frame after alignment are ( The brightness of the pixels; This is the exposure deviation tolerance coefficient.

[0012] Optionally, in step S30, determining the weighted image of the image frame based on the contrast weight, saturation weight, and exposure weight includes: Step S34: According to formula 3-4, the contrast weight, saturation weight, and exposure weight are fused. Formula 3-4 in, The coordinates of the t-th image frame after alignment are ( The fusion weights of the pixels; , , These are the adjustment coefficients for contrast weight, saturation weight, and exposure weight, respectively. Step S35: Normalize the fusion weights according to formula 3-5; Formula 3-5 in, The normalized fusion weights; The constant is T; T is the number of image frames. Step S36: If the translation confidence of the image frame is greater than or equal to the threshold, the weighted image of the aligned image frame is determined directly based on the normalized fusion weights; if the translation confidence of the image frame is less than the threshold, the normalized fusion weights are first proportionally reduced, and then the weighted image of the aligned image frame is determined based on the reduced fusion weights.

[0013] Optionally, step S40 includes: A Laplacian pyramid layered fusion method is used to fuse the various weighted images to obtain a panoramic stitched image.

[0014] The present invention also constructs a processing system for underwater environmental images of nuclear power plants, including a processor and a memory storing a computer program. When the processor executes the computer program, it implements the steps of the above-described processing method for underwater environmental images of nuclear power plants.

[0015] In the technical solution of this application, the translation components of two adjacent image frames are first determined, and each image frame is aligned to the reference coordinate system. Then, the contrast weight, saturation weight, and exposure weight of each aligned image frame are calculated, and the weighted image is determined. Finally, the panoramic stitched image is obtained by fusing the weighted images. Therefore, robust suppression of various random disturbances and robust restoration of moving targets can be achieved, and real details can be highlighted and artifacts can be suppressed, thereby obtaining a clear, stable, and measurable underwater panoramic image, which improves the efficiency and reliability of underwater measurement tasks. Attached Figure Description

[0016] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of a method for processing underwater environment images of a nuclear power plant according to an embodiment of the present invention; Figure 2A This is a schematic diagram of an image frame before processing in one embodiment of the present invention; Figure 2B This is a schematic diagram of a processed image frame in one embodiment of the present invention; Figure 3A This is a schematic diagram of an image frame before processing in one embodiment of the present invention; Figure 3B This is a schematic diagram of a processed image frame in one embodiment of the present invention; Figure 4A This is a schematic diagram of an image frame before processing in one embodiment of the present invention; Figure 4B This is a schematic diagram of a processed image frame in one embodiment of the present invention; Figure 5 This is a schematic diagram of the processed panoramic stitched image in one embodiment of the present invention. Detailed Implementation

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

[0018] Figure 1 This is a flowchart of a method for processing underwater environment images of a nuclear power plant according to an embodiment of the present invention. The processing method of this embodiment includes: Step S10: Convert the acquired video stream of the underwater environment of the nuclear power plant into a sequence of image frames, and preprocess each image frame. In this step, the acquired video stream may be, for example, a video stream of fuel assemblies in a nuclear power plant fuel pool.

[0019] Step S20: Based on the preprocessed image frame sequence, determine the translation components of two adjacent image frames, and align each image frame to the reference coordinate system based on the translation components of two adjacent image frames. Step S30: For each aligned image frame, determine the pixel-level contrast weight, saturation weight, and exposure weight of the image frame, and determine the weighted image of the image frame based on the contrast weight, saturation weight, and exposure weight. In this step, it should be noted that high-contrast areas (edges and textures) in the image frame are usually the most stable and least prone to ghosting in multi-exposure sequences. Therefore, a contrast weight is introduced. To combat underwater "blue-green tint" and color drift caused by thermal disturbance, a saturation weight is introduced to favor more saturated areas during fusion processing. Since the fuel components under test have relatively extreme exposure differences under different fuel consumption, an exposure weight is introduced to achieve sliding attenuation rather than a hard threshold. Pixels closer to the middle exposure have higher quality and are less prone to overexposure or underexposure.

[0020] Step S40: Perform fusion processing on each weighted image to obtain a panoramic stitched image.

[0021] In this embodiment, the acquired video stream is generated under the complex underwater environment of a nuclear power plant (coupling of multiple physical fields such as fluid disturbance, thermal disturbance, scattering, and radiation noise). Therefore, the original image frames suffer from poor imaging quality, large geometric distortion, and high noise. To address this problem, the translation components of two adjacent image frames are first determined, and each image frame is aligned to a reference coordinate system. Then, the contrast weight, saturation weight, and exposure weight of each aligned image frame are calculated, and a weighted image is determined. Finally, a panoramic stitched image is obtained by fusing the weighted images. This approach achieves robust suppression of various random disturbances, robust restoration of moving targets, and highlights realistic details while suppressing artifacts, resulting in a clear, stable, and measurable underwater panoramic image, thus improving the efficiency and reliability of underwater measurement tasks.

[0022] Furthermore, in an optional embodiment, in step S10, converting the acquired video stream of the underwater environment of the nuclear power plant into an image frame sequence includes: decoding the acquired video stream of the underwater environment of the nuclear power plant frame by frame to obtain the image frame sequence. This image frame acquisition method does not require loading the entire video at once, has low memory usage, and also supports real-time processing, that is, performing subsequent processing while decoding.

[0023] Further, in an optional embodiment, step S10 involves preprocessing each image frame, including color correction and / or outlier removal for each image frame. In this embodiment, since underwater scenes or images under multi-exposure environments may exhibit color deviations, color correction is necessary for the image frames. For example, a gray-world-based method can be used for color correction to avoid affecting subsequent weight calculations. Additionally, a local anomaly detection method can be used for outlier removal to effectively suppress CMOS noise.

[0024] Further, in an optional embodiment, in step S20, determining the translation components of two adjacent image frames includes: Step S21: Perform frequency domain phase matching on the grayscale images of two adjacent preprocessed image frames to obtain the cross-correlation surface of the two adjacent image frames. Step S22: Determine the peak value of the cross-correlation surface, and calculate the translation component and translation confidence based on the peak value. It should be noted that the translation component includes horizontal translation and vertical translation. Step S23: For image frames whose translation confidence is less than the threshold, discard the calculated translation component and estimate its translation component relative to the previous image frame using cubic spline interpolation.

[0025] In this embodiment, under multi-physics perturbation, the pixels of a single frame image exhibit random jitter, but the overall change over time can usually be approximated as translation. Therefore, this embodiment uses frequency domain phase correlation to determine the translation components of two adjacent image frames. Moreover, based on translation reliability, unreliable translation components are eliminated, and physical constraints (spatial-temporal consistency) are used for secondary screening and interpolation recovery to re-estimate the translation components. This preserves the motion trend and avoids discontinuous jumps, thereby making the alignment more continuous and smooth.

[0026] Furthermore, in an optional embodiment, in step S21, for each preprocessed image frame, the following is performed: Step S211: Use a two-dimensional window function to perform pixel-level weighting on the grayscale image of the image frame, and perform Fourier transform on the weighted grayscale image to obtain the frequency domain spectrum of the image frame. Step S212: Calculate the normalized cross power spectrum based on the frequency domain spectrum of the image frame and the frequency domain spectrum of the previous image frame, and perform an inverse Fourier transform on the normalized cross power spectrum to obtain the cross-correlation surface between the image frame and the previous image frame.

[0027] In one specific embodiment, step S211 calculates the frequency domain spectrum by pre-setting a two-dimensional window (Hanning window) function. For the t-th image frame after preprocessing, its grayscale image is: .

[0028] First, according to the following formula, use a two-dimensional window function to process the grayscale image. Perform weighted calculations: in, The coordinates of the weighted grayscale image are ( The pixel value of the pixel.

[0029] When performing the above weighting operation, the two-dimensional window function sets pixel-level weights. For example, the pixels in the center of the image have a larger weight, while the pixels in the edge region have a smaller weight. This allows the center of the image to retain a larger signal and the edge region to retain a smaller signal, thus avoiding edge interference in frequency domain analysis.

[0030] Then, according to the following formula, a Fourier transform is performed on the weighted grayscale image to obtain the frequency domain spectrum of the image frame: in, Let be the frequency domain spectrum of the t-th image frame; It is the Fourier transform function; This is the grayscale image after weighted calculation.

[0031] It should be understood that for each preprocessed image frame, the corresponding frequency domain spectrum can be obtained in the manner described above. For example, for the preceding image frame adjacent to the above image frame, its grayscale image is... The grayscale image after weighted operation is Its frequency domain spectrum is .

[0032] In one specific embodiment, in step S212, the normalized cross power spectrum is calculated according to formula 2-1; Formula 2-1 in, Normalized cross-power spectrum; Let be the frequency domain spectrum of the t-th image frame; Let be the frequency domain spectrum of the (t-1)th image frame; It is a constant, such as a small constant used to avoid division by zero.

[0033] Then, calculate the cross-correlation surface between the image frame and its previous image frame according to Formula 2-2; Formula 2-2 in, Let be the cross-correlation surface between the t-th image frame and the (t-1)-th image frame. This is the inverse Fourier transform function.

[0034] Furthermore, after obtaining the cross-correlation surface of two adjacent image frames, the maximum value is found on this cross-correlation surface; this maximum value is the translation component of the two adjacent image frames. Additionally, the translation confidence level can be calculated using the following formula: Formula 2-3 in, Let be the translation confidence level of the t-th image frame; It is the maximum value on the cross-correlation surface between the t-th image frame and the (t-1)-th image frame; It is the average value on the cross-correlation surface between the t-th image frame and the (t-1)-th image frame; To avoid dividing by zero for small constants.

[0035] like If the value is greater than or equal to the threshold, the calculated translation component will be used directly for subsequent calculations.

[0036] like If the value is below a threshold, the image frame is marked as "unreliable". Furthermore, for unreliable image frames, the calculated translation components need to be discarded, and their translation components are estimated using cubic spline interpolation. Specifically, since real motion (camera or object) has the following properties: (1) temporal consistency, i.e., consecutive frames cannot undergo abrupt changes; (2) spatial consistency, i.e., local region displacements should be similar. Values ​​exceeding the threshold are considered unreliable. For unreliable regions, cubic spline interpolation is used to predict the displacement field according to the following formula: in, The translation component of the estimated t-th image frame; These represent the translation components of the k image frames before and after the t-th image frame, respectively, used to fit a smooth curve; This is a cubic spline interpolation function, which fits a smooth curve using the translation components of the 2k adjacent image frames of the t-th image frame and predicts the translation component of the t-th image frame. Moreover, this interpolation algorithm is performed in the time dimension, which can preserve the motion trend and avoid discontinuous jumps.

[0037] Further, in an optional embodiment, in step S20, aligning each image frame to the reference coordinate system according to the translation components of two adjacent image frames includes: Step S24: For each image frame, the translation components of the image frame and each image frame preceding it relative to their respective previous image frame are accumulated to obtain the total translation of the image frame relative to the reference frame. Step S25: For each image frame, construct a translation matrix according to its corresponding total translation amount, and translate the color image of the image frame according to the translation matrix to align the image frame to the reference coordinate system.

[0038] In this embodiment, after obtaining the translation components of two adjacent image frames, the image frames can be translated to align each image frame to the reference coordinate system, that is, to transform to a unified canvas (with a fixed size and a fixed coordinate system), which can eliminate the spatial differences caused by different shooting times and positions.

[0039] In one specific embodiment, after obtaining the translation components of two adjacent image frames, for example, the translation components of the t-th image frame and the (t-1)-th image frame... It can be represented as: in, The translation component of the t-th image frame relative to the (t-1)-th image frame along the x-axis (horizontal direction); The translation component of the t-th image frame relative to the (t-1)-th image frame along the y-axis (vertical direction).

[0040] To transform image frames onto a uniform canvas, the total translation relative to the reference frame for each image frame can be calculated using the following formula: The reference frame can be the 0th image frame; The total translation of the t-th image frame relative to the reference frame along the x-axis; The total translation of the t-th image frame relative to the reference frame along the y-axis; The translation component on the x-axis of the i-th image frame compared to the (i-1)-th image frame; It represents the translation component of the i-th image frame on the y-axis compared to the (i-1)-th image frame.

[0041] Next, based on the determined total translation, construct the following 2×3 translation matrix: Then, for the preprocessed image frame, a translation transformation is performed on it according to the translation matrix, which can be expressed as: in,( , () represents the coordinates before translation, ( , ) represents the coordinates after translation.

[0042] Further, in an optional embodiment, in step S30, determining the contrast weight, saturation weight, and exposure weight at the pixel level of the image frame includes: Step S31: Calculate the pixel-level contrast weights of the image frame according to Formula 3-1; Formula 3-1 in, The coordinates of the t-th image frame after alignment are ( The contrast weight of the pixels; This refers to the t-th image frame after alignment. This is a grayscale conversion function used to extract the luminance component; This represents the Laplace operator, which is a second-order derivative operator used to detect local second-order changes. Specifically, , The second derivative is the result of the Laplacian operation, which represents the second derivative of the gray-level changes around the pixel. The higher the value, the more drastic the gray-level changes near the pixel, corresponding to edges or textures. Then, the absolute value at (x, y) is calculated based on the second derivative result.

[0043] Step S32: Calculate the pixel-level saturation weight of the image frame according to Formula 3-2; Formula 3-2 in, The coordinates of the t-th image frame after alignment are ( The saturation weight of the pixels; The coordinates of the t-th image frame after alignment are ( The pixel value of the pixel in channel c (where c is R, G, B). The coordinates of the t-th image frame are ( The average pixel value of a pixel across the R, G, and B channels; Step S33: Calculate the pixel-level exposure weight of the image frame according to formula 3-3; Formula 3-3 in, The coordinates of the t-th image frame after alignment are ( Exposure weights of pixels; It is a Gaussian function; The coordinates of the t-th image frame after alignment are ( The brightness (Y channel) of the pixel; This is the exposure deviation tolerance factor, for example, 0.2 to 0.3; 0.5 represents the mid-gray value of the ideal exposure (after normalization).

[0044] Further, in an optional embodiment, in step S30, determining the weighted image of the image frame based on the contrast weight, saturation weight, and exposure weight includes: Step S34: According to formula 3-4, the contrast weight, saturation weight, and exposure weight are fused. Formula 3-4 in, The coordinates of the t-th image frame after alignment are ( The fusion weights of the pixels; , , These are the adjustment coefficients for contrast weight, saturation weight, and exposure weight, respectively. Step S35: Normalize the fusion weights according to formula 3-5; Formula 3-5 in, The normalized fusion weights; It is a constant, for example, a small constant used to avoid division by zero; T is the number of image frames; Step S36: If the translation confidence of the image frame is greater than or equal to the threshold, the weighted image of the aligned image frame is determined directly based on the normalized fusion weights; if the translation confidence of the image frame is less than the threshold, the normalized fusion weights are first proportionally reduced, and then the weighted image of the aligned image frame is determined based on the reduced fusion weights.

[0045] In this embodiment, the fusion weight is determined by weighted summation of contrast weight, saturation weight, and exposure weight. Compared to the commonly used "product weight," this method is more stable and avoids excessively sparse weights when any one factor is small. Furthermore, the influence of translation confidence is introduced into the fusion weight. That is, for image frames with low translation confidence, the calculated fusion weight is reduced proportionally. For example, the calculated fusion weight is multiplied by a coefficient less than 1. The value of this coefficient depends on the translation confidence; the lower the translation confidence, the smaller the coefficient. Thus, for image frames with low translation confidence, the larger the reduction ratio, the smaller the translation contribution, thereby avoiding unreliable image frames from having an excessive impact on global alignment.

[0046] Further, in an optional embodiment, step S40 includes: using a Laplacian pyramid layered fusion method to fuse the various weighted images to obtain a panoramic stitched image. In this embodiment, the advantage of using the Laplacian pyramid layered fusion method is that low-frequency content is controlled by large-scale weights (smooth transition), while high-frequency content is controlled by local weights (edge ​​and detail preservation), resulting in a smooth, seamless fusion result that preserves details, greatly reducing seams and ghosting. Moreover, during this process, flicker noise in high frequencies is quickly eliminated, while relatively stable target structures in low frequencies can be stably reconstructed, and local abnormal disturbances do not spread across layers.

[0047] The present invention also constructs a processing system for underwater environmental images of nuclear power plants. The processing system includes a processor and a memory storing a computer program. When the processor executes the computer program, it implements the steps of the above-described processing method for underwater environmental images of nuclear power plants.

[0048] In summary, to address the following technical issues: In the complex underwater environment of nuclear power plants (multi-physics coupling including fluid disturbance, thermal disturbance, scattering, and radiation noise), single-frame images suffer from poor image quality, large geometric distortion, high noise, and low measurement accuracy. For example, Figure 2A , Figure 3A The image frame shown contains perturbations. Figure 4A The image frames shown contain noise. Existing technologies cannot simultaneously address the issues of "scattering + thermal disturbance + radiation noise" under multi-physics coupling conditions, thus requiring manual compensation or repeated shooting for critical underwater measurement tasks, impacting efficiency and reliability. This invention constructs a mathematical model for improving imaging quality based on multi-frame image stabilization, reliability weighting, interpolation compensation, and pyramid hierarchical fusion. This model achieves robust suppression of random disturbances, robust restoration of moving targets, and reliable estimation of featureless regions. For example, Figure 2B , Figure 3B The image frames shown have been de-disturbed. Figure 4B The image frames shown have been noise-removed to obtain clear, stable, and measurable underwater panoramic images, such as... Figure 5 As shown.

[0049] 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 the claims of this application.

Claims

1. A method for processing underwater environment images of a nuclear power plant, characterized in that, include: Step S10: Convert the acquired video stream of the underwater environment of the nuclear power plant into a sequence of image frames, and preprocess each image frame. Step S20: Based on the preprocessed image frame sequence, determine the translation components of two adjacent image frames, and align each image frame to the reference coordinate system based on the translation components of two adjacent image frames. Step S30: For each aligned image frame, determine the pixel-level contrast weight, saturation weight, and exposure weight of the image frame, and determine the weighted image of the image frame based on the contrast weight, saturation weight, and exposure weight. Step S40: Perform fusion processing on each weighted image to obtain a panoramic stitched image.

2. The method for processing underwater environment images of nuclear power plants according to claim 1, characterized in that, In step S10, the preprocessing of each image frame includes: For each image frame, perform color correction and / or outlier removal.

3. The method for processing underwater environment images of nuclear power plants according to claim 1, characterized in that, In step S20, determining the translation components of two adjacent image frames includes: Step S21: Perform frequency domain phase matching on the grayscale images of two adjacent preprocessed image frames to obtain the cross-correlation surface of the two adjacent image frames. Step S22: Determine the peak value of the cross-correlation surface, and calculate the translation component and translation confidence level based on the peak value; Step S23: For image frames whose translation confidence is less than the threshold, discard the calculated translation component and estimate its translation component relative to the previous image frame using cubic spline interpolation.

4. The method for processing underwater environment images of nuclear power plants according to claim 3, characterized in that, In step S21, for each preprocessed image frame, the following is performed: Step S211: Use a two-dimensional window function to perform pixel-level weighting on the grayscale image of the image frame, and perform Fourier transform on the weighted grayscale image to obtain the frequency domain spectrum of the image frame. Step S212: Calculate the normalized cross power spectrum based on the frequency domain spectrum of the image frame and the frequency domain spectrum of the previous image frame, and perform an inverse Fourier transform on the normalized cross power spectrum to obtain the cross-correlation surface between the image frame and the previous image frame.

5. The method for processing underwater environment images of nuclear power plants according to claim 4, characterized in that, In step S212, the normalized cross power spectrum is calculated according to formula 2-1; Official 2-1 in, Normalized cross-power spectrum; Let be the frequency domain spectrum of the t-th image frame; Let be the frequency domain spectrum of the (t-1)th image frame; It is a constant.

6. The method for processing underwater environment images of nuclear power plants according to claim 1, characterized in that, In step S20, aligning each image frame to the reference coordinate system based on the translation components of two adjacent image frames includes: Step S24: For each image frame, the translation components of the image frame and each image frame preceding it relative to their respective previous image frame are accumulated to obtain the total translation of the image frame relative to the reference frame. Step S25: For each image frame, construct a translation matrix according to its corresponding total translation amount, and translate the color image of the image frame according to the translation matrix to align the image frame to the reference coordinate system.

7. The method for processing underwater environment images of nuclear power plants according to claim 3, characterized in that, In step S30, determining the contrast weight, saturation weight, and exposure weight at the pixel level of the image frame includes: Step S31: Calculate the pixel-level contrast weights of the image frame according to Formula 3-1; Official 3-1 in, The coordinates of the t-th image frame after alignment are ( The contrast weight of the pixels; This refers to the t-th image frame after alignment. This is a grayscale conversion function; For the Laplace operator; Step S32: Calculate the pixel-level saturation weight of the image frame according to Formula 3-2; Official 3-2 in, The coordinates of the t-th image frame after alignment are ( The saturation weight of the pixels; The coordinates of the t-th image frame after alignment are ( The pixel value of the pixel in channel c. The coordinates of the t-th image frame are ( The average pixel value of a pixel across the three channels; Step S33: Calculate the pixel-level exposure weight of the image frame according to formula 3-3; Official 3-3 in, The coordinates of the t-th image frame after alignment are ( Exposure weights of pixels; It is a Gaussian function; The coordinates of the t-th image frame after alignment are ( The brightness of the pixels; This is the exposure deviation tolerance coefficient.

8. The method for processing underwater environment images of a nuclear power plant according to claim 7, characterized in that, In step S30, the weighted image of the image frame is determined based on the contrast weight, saturation weight, and exposure weight, including: Step S34: According to formula 3-4, the contrast weight, saturation weight, and exposure weight are fused. Official 3-4 in, The coordinates of the t-th image frame after alignment are ( The fusion weights of the pixels; , , These are the adjustment coefficients for contrast weight, saturation weight, and exposure weight, respectively. Step S35: Normalize the fusion weights according to formula 3-5; Official 3-5 in, The normalized fusion weights; The constant is T; T is the number of image frames. Step S36: If the translation confidence of the image frame is greater than or equal to the threshold, the weighted image of the aligned image frame is determined directly based on the normalized fusion weights; if the translation confidence of the image frame is less than the threshold, the normalized fusion weights are first proportionally reduced, and then the weighted image of the aligned image frame is determined based on the reduced fusion weights.

9. The method for processing underwater environment images of a nuclear power plant according to claim 1, characterized in that, Step S40 includes: A Laplacian pyramid layered fusion method is used to fuse the various weighted images to obtain a panoramic stitched image.

10. A system for processing underwater environmental images of a nuclear power plant, comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for processing underwater environment images of nuclear power plants according to any one of claims 1-9.

Images