A sequence image fusion enhancement method based on three-axis slide table displacement sensing

Abstract

This invention discloses a sequential image fusion enhancement method based on three-axis slide table displacement sensing, belonging to the field of machine vision technology. The method includes the following steps: S1, constructing discretized state and output equations to complete the three-axis slide table motion modeling, and establishing the correlation between motion parameters and image blur kernel and pixel position offset; S2, using the inverse function of the camera response function combined with Taylor linearization, performing exposure normalization processing on multiple frames of observed images to obtain scene illumination estimates with a unified radiance scale; S3, obtaining multi-dimensional fusion weights for the image by setting a fusion weight calculation mechanism that correlates the quality features of the fused image with the slide table position; S4, completing the fusion of multiple frames of images through a weighted fusion model to obtain a high-quality single-frame fused image. Using the above method, the problems of image motion blur and uneven multi-frame imaging quality caused by three-axis slide table motion are effectively solved, achieving accurate fusion and deblurring enhancement of sequential images.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to a sequential image fusion enhancement method based on triaxial slide displacement sensing. Background Technology

[0002] In existing technologies, motion blur removal methods mainly include traditional blind deconvolution algorithms and machine learning-based methods. The former, such as methods based on natural image priors, the Richardson-Lucy algorithm and its variants, achieves blur kernel estimation and image restoration by assuming prior image characteristics or iterative optimization, and shows certain effectiveness in static scenes or when the blur kernel is known. The latter, through various technical paths such as blur kernel estimation, multi-scale network construction, and recurrent residual network improvement, improves the deblurring effect in specific scenarios, providing diversified solutions for motion blur removal problems.

[0003] However, traditional blind deconvolution algorithms are sensitive to initial values, prone to artifacts or detail distortion, and have high computational complexity, making them difficult to meet real-time requirements. Machine learning-based methods rely on large-scale labeled data, have insufficient generalization ability, and poor adaptability. More importantly, these methods do not fully incorporate the motion characteristics of the three-axis slide in the automated system, ignore the inherent relationship between motion parameters and image blur, and cannot achieve efficient deblurring when faced with significant differences in brightness, sharpness, and noise among multiple frames of images, as well as core problems such as the lack of correlation between motion patterns and images and a single fusion weight design. As a result, the image quality in complex motion blur scenarios is difficult to improve effectively. Summary of the Invention

[0004] The purpose of this invention is to provide a sequential image fusion enhancement method based on triaxial slide displacement sensing, thereby solving the above-mentioned technical problems.

[0005] To achieve the above objectives, this invention provides a sequential image fusion enhancement method based on triaxial slide displacement sensing, comprising the following steps: S1. Using the three-axis motor control command of the three-axis slide as input parameters, construct discretized state equations and output equations to complete the motion modeling of the three-axis slide, obtain the motion parameters of the slide in the X, Y, and Z axes, and establish the correlation between the motion parameters and the image blur kernel and pixel position offset. S2. Based on the motion parameters in S1, the exposure normalization process is performed on the multi-frame observation images by using the inverse function of the camera response function combined with Taylor linearization to obtain scene illumination estimates with a uniform radiance scale. S3. Based on the motion parameters of S1 and the scene illumination estimation value of S2, the multi-dimensional fusion weight of the image is obtained by setting the fusion weight calculation mechanism of the correlation between the fused image quality features and the slide position. S4, based on the scene illumination estimation value of S2 and the multi-dimensional fusion weight of S3, completes the fusion of multiple frames of images through a weighted fusion model to obtain a high-quality single-frame fused image.

[0006] Preferably, the specific steps in S1 for constructing discretized state equations and output equations to complete the three-axis slide table motion modeling include: S11. Based on the relative assembly relationship between the slide and the imaging camera, a three-dimensional rectangular coordinate system is defined with the initial position of the slide as the origin and along the image dimension and the optical axis, to obtain the three-axis motion reference direction adapted to the image pixel scale. S12. Based on the triaxial motion reference direction determined in S11 and the actual working condition of the slide, the standardized dynamic boundary conditions for adaptive modeling are obtained by applying triaxial motion decoupling constraints, linear dynamic constraints, fixed sampling period discrete constraints and rigid connection constraints between the slide and the camera. S13. Based on the standardized dynamic boundary conditions obtained in S12, a unified parameter system required to construct the equations of motion is obtained by defining a parameter set including system input, mechanical parameters, state vector and output vector. S14. Based on the unified parameter system obtained in S13 and the triaxial load dynamic equations established based on Newton's second law, and combined with the first-order Euler method to discretize the continuous domain dynamic equations, a discretized state equation is constructed, the formula of which is: ; in, For the first The system state vector for each control cycle; This is the state transition matrix; The input matrix; S15. Based on the discretized state equations obtained in S14, and taking the three-dimensional spatial displacement at the end of the sliding table as the system output, construct the output equation, which is: ; in, This is the output matrix; S16, based on the discretized state equation of S14 and the output equation of S15, through matrix invertibility, system stability and pixel-level error verification, an effective three-axis slide table motion model that meets engineering accuracy is obtained.

[0007] Preferably, the parameter set in S13 specifically includes: the three-axis motor input vector. Motor thrust / torque constant matrix Three-axis equivalent mass matrix Viscous damping coefficient matrix Three-axis velocity vector Three-axis position vector System state vector and system output vector ; Wherein, the system state vector For containing three-axis velocity vectors With the three-axis position vector A 6×1 dimensional vector, the system output vector It is a 3×1 dimensional vector representing the three-dimensional spatial displacement of the end of the slide.

[0008] Preferably, the specific steps in S1 for establishing the correlation between the three-dimensional motion parameters and the image blur kernel and pixel position offset include: S17. Calculate the spatial displacement of the slide during the exposure time based on the three-dimensional motion parameters, and convert the spatial displacement into pixel displacement by combining the camera calibration parameters. Construct a motion blur kernel for the two-dimensional pixel plane, using the following formula: ; in, For the first The motion blur kernel function corresponding to the frame image; To obscure the extent of nuclear diffusion, and , , For the spatial displacement of the sliding table, , These are the calibration coefficients from spatial displacement to pixel displacement; S18, Constructing the first based on motion fuzzy kernel A motion imaging model for frame images to clearly define the correlation between pixel position offset and pixel point. The imaging formula is: ; in, The grayscale value of the image; The actual lighting intensity of the scene; For the first The exposure time of the frame; The motion fuzzy kernel for the sliding table; It is noise.

[0009] Preferably, in S18, based on the fact that the slide table maintains motion during the acquisition process, the pixel points caused by the slide table motion are... The set of location change observations is as follows: ; in, , For the first Pixels in a frame image Relative to the pixels in the first frame image Pixel displacement in the X and Y directions.

[0010] Preferably, the specific steps of S2 include: S21. Calibrate the camera's response function using a standard grayscale plate experiment. And using the inverse function of the camera response function Calculate the first The scene lighting corresponding to the frame, the first Frame image is The calculation formula is: ; in, For the first Within a frame, the estimated scene physical radiosity value corresponding to the pixel position after displacement; For the first In a frame, the reference pixel coordinates The original estimated value of the scene's physical radiosity corresponding to the actual pixel position after displacement by the sliding table; S22. Based on the pixel displacement generated by the sliding table maintaining its motion state during the acquisition process, Taylor linearization is used to establish a first-order approximation relationship between the current pixel and the displaced pixel to achieve image alignment and normalization. The formula for the first-order approximation is: ; in, For the first In a frame, the reference pixel coordinates Approximate value of scene physical radiosity estimate corresponding to the pixel position after displacement by the sliding table; For the first In the frame, the reference pixel coordinates do not take into account the slide displacement. The corresponding scene physical radiosity estimation baseline value; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; S23. Based on the first-order approximation of S22, perform rearrangement and transformation, and combine with the original S21. The formula for calculating exposure normalization is obtained to calculate the first... The normalized radiance estimate for the scene illumination in a frame is given by the following formula: ; in, For the first In a frame, the reference pixel coordinates The final estimated value of the scene's physical radiometric intensity after exposure normalization correction.

[0011] Preferably, the formula for the multi-dimensional fusion weights constructed in S3 is: ; in, For multi-dimensional weight fusion; , , , These are the weighting coefficients; For clarity weight; Brightness weight; Noise weights; For positional weights.

[0012] Preferably, the sharpness weight in S3 The Laplacian operator is used to calculate the edge intensity of the image. The calculation formula is as follows: ; in, For the Laplace operator; For the first In the frame image, the coordinates are The original grayscale value corresponding to the pixel; This represents the total number of frames in the multi-frame observation image; Brightness weight The calculation formula is: ; in, , For a reasonable brightness range; Noise weight The signal-to-noise ratio (SNR) of the image is determined by calculation, using the following formula: ; in, For the first Signal-to-noise ratio of a frame image; Position weight The calculation formula is: ; in, For the first The distance between a frame and a target frame; the smaller the distance, the stronger the positional correlation, and the greater the weight. .

[0013] Preferably, the specific steps of S4 include: S41. The normalized result obtained in S23 And S3 A weighted summation model is used to achieve multi-frame fusion, resulting in the optimal illumination estimate for the target frame. The formula is: ; S42, via camera response function S41 Map back to grayscale to obtain a high-quality single-frame fused image. The calculation formula is: .

[0014] Therefore, the present invention employs the above-mentioned sequential image fusion enhancement method based on triaxial slide displacement sensing, which has the following beneficial effects: 1. By defining a three-dimensional Cartesian coordinate system adapted to the pixel scale of the image, applying standardized dynamic boundary conditions, constructing discretized state equations and output equations, a three-axis slide table motion model that meets engineering accuracy is formed. The correlation between motion parameters and image blur kernel and pixel position offset is established. At the same time, the pixel displacement observation set of multiple frames of images is clarified, breaking through the limitation of existing technology that ignores the motion characteristics of the equipment. The intrinsic relationship between motion law and imaging quality is clearly established, providing accurate kinematic basis for subsequent image alignment and blur correction, and solving the core problems such as pixel position offset and motion blur caused by slide table motion.

[0015] 2. Based on the sliding table motion parameters, the camera response function is calibrated through a standard grayscale test. The original estimate of the scene physical radiometer of the pixel position after displacement is calculated using its inverse function. Then, a first-order approximation relationship of pixel displacement is established through Taylor linearization. After shifting terms, the exposure normalization correction formula is obtained, and the final estimate of the scene physical radiometer at a unified radiometer scale is output. This eliminates the radiometer estimation deviation caused by differences in different exposure times and sliding table pixel displacement, realizes the unification of multiple frames of images at the physical radiometer scale, avoids the inaccurate correction problem caused by traditional exposure normalization not considering motion displacement, and improves the brightness uniformity of the fused image.

[0016] 3. Based on the sliding table motion parameters and the normalized scene illumination estimates, a multi-dimensional fusion weight calculation mechanism is constructed, which includes sharpness weight, brightness weight, noise weight, and position weight. Each component weight is obtained through the Laplacian operator, reasonable brightness range determination, signal-to-noise ratio calculation, and sliding table position distance calculation, respectively. The proportion of each component weight is adjusted by calibration coefficients. This comprehensively considers the image's own quality characteristics and the positional correlation brought about by the sliding table motion, allowing sharp frames, frames with reasonable brightness, low noise frames, and frames adjacent to the sliding table to play a greater role in the fusion process. This avoids fusion deviation caused by a single weight and improves the detail integrity and noise resistance of the fusion result.

[0017] 4. By substituting the normalized final estimate of the scene's physical radiometrics with the multi-dimensional fusion weights into the weighted summation model, the optimal illumination estimate of the target frame is obtained. Then, the camera response function is used to map back to the grayscale scale, outputting a high-quality single-frame fused image. This fully utilizes the complementary information of multiple frames, allowing blurred frames to provide low-frequency brightness information and clear frames to supplement high-frequency detail information, achieving a synergistic effect of motion blur removal, noise suppression, and brightness optimization. Ultimately, a fused image with high clarity, uniform brightness, and strong anti-interference capability is obtained, meeting the high-precision image quality requirements of key links such as target recognition, size measurement, and defect detection in automated systems.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 The flowchart illustrates a sequential image fusion enhancement method based on triaxial slide displacement sensing provided by this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0021] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0022] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Against the backdrop of the deep integration of automation technology and image processing technology, three-axis sliding table dynamic shooting scenarios generally suffer from problems such as uneven image brightness, motion blur, and uneven noise distribution. Existing motion blur removal methods do not fully incorporate the motion characteristics of the sliding table and cannot solve the core pain points of significant differences between multiple frames, lack of correlation between motion patterns and images, and simplistic fusion weight design.

[0024] Based on the above analysis, this invention is designed, see appendix. Figure 1 A sequence image fusion enhancement method based on triaxial slide displacement sensing includes the following steps: S1. Using the three-axis motor control command of the three-axis slide as input parameters, construct discretized state equations and output equations to complete the motion modeling of the three-axis slide, obtain the motion parameters of the slide in the X, Y, and Z axes, and establish the correlation between the motion parameters and the image blur kernel and pixel position offset. The specific steps of S1 include: S11. Based on the relative assembly relationship between the slide and the imaging camera, a three-dimensional rectangular coordinate system is defined with the initial position of the slide as the origin and along the image dimension and the optical axis, to obtain the three-axis motion reference direction adapted to the image pixel scale. Specifically defined as follows: The X-axis is with the initial position of the slide as the origin, and is parallel to the width direction of the image in the horizontal direction. The positive direction of the X-axis is to the right, and the negative direction of the X-axis is to the left. The Y-axis is defined with the initial position of the slide as the origin, and is parallel to the image height direction in a vertical direction. The positive Y-axis is upward, and the negative Y-axis is downward. The Z-axis is defined by taking the initial position of the slide as the origin and extending along a direction perpendicular to the image plane, with the positive Z-axis pointing inwards and the negative Z-axis pointing outwards.

[0025] S12. Based on the triaxial motion reference direction determined in S11 and the actual working condition of the slide, the standardized dynamic boundary conditions for adaptive modeling are obtained by applying triaxial motion decoupling constraints, linear dynamic constraints, fixed sampling period discrete constraints and rigid connection constraints between the slide and the camera. The three-axis motion decoupling assumption is that the X, Y, and Z axis motions are independent of each other and have no mechanical coupling interference. The dynamic characteristics of each axis can be modeled separately and then integrated by matrix. The linear dynamics assumption is that each axis satisfies the dynamic characteristics of a linear second-order system. Parameters such as motor driving force, load inertia, and damping coefficient are constants and can be obtained through experimental calibration or by consulting the manual. The discretization assumption is that a fixed sampling period is used to discretize the continuous motion process, and the exposure time and discretization error are negligible. The rigid connection is assumed to be a rigid connection between the slide, the camera, and the load, with no relative deformation, and the displacement at the end of the slide is equal to the displacement of the camera.

[0026] S13. Based on the standardized dynamic boundary conditions obtained in S12, a unified parameter system required to construct the equations of motion is obtained by defining a parameter set including system input, mechanical parameters, state vector and output vector. The parameter set specifically includes: the input vector of the three-axis motor. Motor thrust / torque constant matrix Three-axis equivalent mass matrix Viscous damping coefficient matrix Three-axis velocity vector Three-axis position vector System state vector and system output vector ; Wherein, the system state vector For containing three-axis velocity vectors With the three-axis position vector A 6×1 dimensional vector, the system output vector It is a 3×1 dimensional vector representing the three-dimensional spatial displacement of the end of the slide.

[0027] See Table 1 for details: Table 1 Definition of Core Parameters

[0028] Based on the electromagnetic torque / thrust characteristics of the motor, the output driving force vector of the three-axis motor satisfies a linear mapping relationship with the input vector. A motor driving force model is constructed, and its matrix expansion is as follows: ; in, This is the driving force vector output by the three-axis motor. This model reflects the conversion relationship between the motor input signal and the mechanical driving force.

[0029] S14. Based on the unified parameter system obtained in S13 and the triaxial load dynamic equations established based on Newton's second law, and combined with the first-order Euler method to discretize the continuous domain dynamic equations, a discretized state equation is constructed, the formula of which is: ; in, For the first The system state vector for each control cycle; Let be the state transition matrix, and ; Given an input matrix, and ; The derivation of the triaxial load dynamic equation is as follows: Based on Newton's second law, the dynamic behavior of the single-axis slide table satisfies a second-order linear differential equation, the formula of which is: ; Integrating the triaxial dynamic equations into a matrix expansion, we get: ; in, Let be the three-axis acceleration vector, and the differential relationship between position and velocity is: ; The derivation process of the discretized state equation is as follows: To meet the demands of real-time computer computing and control, the first-order Euler method is used to discretize the continuous-domain dynamic equations, with a sampling period of [missing information]. Based on the kinematic equations, the expansion formula of the acceleration matrix after velocity discretization and updating is: ; in, ; Substituting the acceleration matrix expansion formula into the matrix expansion of the motor driving force model, we get: ; Combining the first-order Euler discretization formula, the three-axis velocity update equations are obtained by expansion: ; After sorting and breaking down, the velocity update expressions for each axis are obtained as follows: ; The location discretization process is as follows: Based on the differential relationship between position and velocity, the three-axis position update equation is obtained by expanding using the first-order Euler discretization formula: ; The expression for updating the position of each axis is obtained by splitting the expression: ; velocity vector With position vector Integrate into system state vector And combined with the discretized state equations, the state transition matrix is... and input matrix Substituting the values, we get the complete matrix as follows: ; This equation fully describes the evolution of the system state, including three-axis velocity and position, with control input. The matrix expansion form intuitively presents the coupled influence of each parameter on the motion state.

[0030] S15. Based on the discretized state equations obtained in S14, and taking the three-dimensional spatial displacement at the end of the sliding table as the system output, construct the output equation, which is: ; in, This is the output matrix; The derivation process of the output equation is as follows: The core output of the system is the three-dimensional spatial displacement of the end of the slide, i.e., the position component in the state vector. Therefore, the matrix expansion of the output equation is as follows: ; Among them, the output matrix The expansion form clarifies the mapping relationship for extracting 3D position information from a 6D state vector, and directly outputs the 3D displacement of the slide end. This provides a quantitative basis for subsequent image position offset correction and motion blur kernel estimation.

[0031] S16, based on the discretized state equation of S14 and the output equation of S15, through matrix invertibility, system stability and pixel-level error verification, an effective three-axis slide table motion model that meets engineering accuracy is obtained. The matrix invertibility is verified as follows: It is a diagonal matrix and its diagonal elements , , All are greater than 0, therefore, its inverse matrix is It exists and is unique; System stability is verified by: state transition matrix eigenvalues , , , The parameters need to be calibrated to ensure that the magnitude of the first three eigenvalues ​​is less than 1 in order to ensure the stability of the system after discretization. Pixel-level error verification is performed as follows: the mean square error (MSE) between the model-predicted displacement and the actual displacement satisfies: ; in, To predict displacement; This represents the actual displacement to meet the pixel-level positioning requirements of the image.

[0032] S17. Calculate the spatial displacement of the slide during the exposure time based on the three-dimensional motion parameters, and convert the spatial displacement into pixel displacement by combining the camera calibration parameters. Construct a motion blur kernel for the two-dimensional pixel plane, using the following formula: ; in, For the first The motion blur kernel function corresponding to the frame image; To obscure the extent of nuclear diffusion, and , , For the spatial displacement of the sliding table, , These are the calibration coefficients from spatial displacement to pixel displacement; When the exposure time of the industrial camera is much shorter than the macroscopic motion period of the slide table, the slide table acceleration term can be considered constant during the exposure time, and its position change is approximately as follows: ; in, For the first The start time of exposure during frame image acquisition; For the first The exposure time of the industrial camera corresponding to the frame image; This represents the initial position of the slide on the X-axis. The start time The speed of the slide along the X-axis; available , ; The velocity is obtained by updating the velocity of each axis based on the decomposition. and ; , The calibration coefficients for the conversion from spatial displacement to pixel displacement are obtained through camera calibration and are as follows: , ; in, , The physical size corresponding to each pixel of the camera; The pixel displacement is: , ; Therefore, the pixel scale of the blur kernel is obtained as follows: ; Substituting, we get: .

[0033] S18, Constructing the first based on motion fuzzy kernel A motion imaging model for frame images to clearly define the correlation between pixel position offset and pixel point. The imaging formula is: ; in, The grayscale value of the image; The actual light intensity of the scene over time The changes reflect the physical brightness of the surface of the photographed object; For the first The exposure time of the frame; The motion blur kernel of the slide table is determined by the motion law of the slide table and describes the blurring effect of the motion trajectory of the object being photographed during the exposure time on the image. Noise, such as random interference introduced by the camera sensor or environmental disturbances; In S18, based on the fact that the slide table remains in motion during the acquisition process, the pixel points are determined according to the movement of the slide table. The set of location change observations is as follows: ; in, , For the first Pixels in a frame image Relative to the pixels in the first frame image Pixel displacement in the X and Y directions.

[0034] S2. Based on the motion parameters in S1, the exposure normalization process is performed on the multi-frame observation images by using the inverse function of the camera response function combined with Taylor linearization to obtain scene illumination estimates with a uniform radiance scale. The specific steps of S2 include: S21. Calibrate the camera's response function using a standard grayscale plate experiment. And using the inverse function of the camera response function Calculate the first The scene lighting corresponding to the frame, the first Frame image is The calculation formula is: ; in, For the first Within a frame, the estimated scene physical radiosity value corresponding to the pixel position after displacement; For the first In a frame, the reference pixel coordinates The original estimated value of the scene's physical radiosity corresponding to the actual pixel position after displacement by the sliding table; S22. Based on the pixel displacement generated by the sliding table maintaining its motion state during the acquisition process, Taylor linearization is used to establish a first-order approximation relationship between the current pixel and the displaced pixel to achieve image alignment and normalization. The formula for the first-order approximation is: ; in, For the first In a frame, the reference pixel coordinates Approximate value of scene physical radiosity estimate corresponding to the pixel position after displacement by the sliding table; For the first In the frame, the reference pixel coordinates do not take into account the slide displacement. The corresponding scene physical radiosity estimation baseline value; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; S23. Based on the first-order approximation of S22, perform rearrangement and transformation, and combine with the original S21. The formula for calculating exposure normalization is obtained to calculate the first... The normalized radiance estimate for the scene illumination in a frame is given by the following formula: ; in, For the first In a frame, the reference pixel coordinates The final estimated value of the scene's physical radiometric intensity after exposure normalization correction.

[0035] This linear approximation holds true under the condition that the exposure time is much shorter than the slide movement period and the brightness field is continuously differentiable, and can be used to quickly calculate exposure-compensated images.

[0036] S3. Based on the motion parameters of S1 and the scene illumination estimation value of S2, the multi-dimensional fusion weight of the image is obtained by setting the fusion weight calculation mechanism of the correlation between the fused image quality features and the slide position. The formula for the multi-dimensional fusion weights constructed in S3 is: ; in, For multi-dimensional weight fusion; , , , These are weighting coefficients, calibrated experimentally. For clarity weight; Brightness weight; Noise weights; Position weights; Sharpness weight in S3 The Laplacian operator is used to calculate the edge intensity of an image. The greater the edge intensity, the clearer the image, and the greater its weight. The calculation formula is as follows: ; in, This is the Laplacian operator, used to calculate the edge intensity of image pixels; For the first In the frame image, the coordinates are The original grayscale value corresponding to the pixel; This represents the total number of frames in the multi-frame observation image; Brightness weight The calculation formula is: ; in, , The range is considered reasonable; when the brightness exceeds this range, the weight decreases linearly with the degree of deviation. Noise weight The signal-to-noise ratio (SNR) of the image is determined by calculation. A higher SNR indicates less noise interference and a greater weighting. The formula is as follows: ; in, For the first Signal-to-noise ratio of a frame image; Position weight The calculation formula is: ; in, For the first The distance between a frame and a target frame; the smaller the distance, the stronger the positional correlation, and the greater the weight. .

[0037] S4, based on the scene illumination estimation value of S2 and the multi-dimensional fusion weight of S3, multi-frame image fusion is completed through a weighted fusion model to obtain a high-quality single-frame fused image; The specific steps of S4 include: S41. The normalized result obtained in S23 And S3 A weighted summation model is used to achieve multi-frame fusion, resulting in the optimal illumination estimate for the target frame. The formula is: ; S42, via camera response function S41 Map back to grayscale to obtain a high-quality single-frame fused image. The calculation formula is: .

[0038] In summary, this invention addresses the problems of uneven brightness, sharpness differences, and uneven noise distribution in multi-frame images under three-axis slide table motion scenarios, as well as the shortcomings of existing motion blur removal methods, such as getting trapped in local optima, relying on labeled data, and failing to fully incorporate slide table motion characteristics, resulting in poor fusion effects. By constructing a technical process of "motion modeling—exposure normalization—weighted fusion," this invention accurately quantifies the correlation between slide table motion parameters and image blur, unifies the radiometric scale of multi-frame images, and designs multi-dimensional fusion weights that consider both image quality and slide table position correlation. This effectively eliminates motion blur, brightness deviation, and noise interference, significantly improving the sharpness, brightness uniformity, and anti-interference capability of the fused image. It provides high-precision and reliable image support for key aspects of automated systems such as target recognition, size measurement, and defect detection.

[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A sequence image fusion and enhancement method based on triaxial slide table displacement sensing, characterized in that: Includes the following steps: S1. Using the three-axis motor control command of the three-axis slide as input parameters, construct discretized state equations and output equations to complete the motion modeling of the three-axis slide, obtain the motion parameters of the slide in the X, Y, and Z axes, and establish the correlation between the motion parameters and the image blur kernel and pixel position offset. S2. Based on the motion parameters in S1, the exposure normalization process is performed on the multi-frame observation images by using the inverse function of the camera response function combined with Taylor linearization to obtain scene illumination estimates with a uniform radiance scale. S3. Based on the motion parameters of S1 and the scene illumination estimation value of S2, the multi-dimensional fusion weight of the image is obtained by setting the fusion weight calculation mechanism of the correlation between the fused image quality features and the slide position. S4, based on the scene illumination estimation value of S2 and the multi-dimensional fusion weight of S3, completes the fusion of multiple frames of images through a weighted fusion model to obtain a high-quality single-frame fused image.

2. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 1, characterized in that: The specific steps in S1 to construct discretized state and output equations to complete the three-axis slide table motion modeling include: S11. Based on the relative assembly relationship between the slide and the imaging camera, a three-dimensional rectangular coordinate system is defined with the initial position of the slide as the origin and along the image dimension and the optical axis, to obtain the three-axis motion reference direction adapted to the image pixel scale. S12. Based on the triaxial motion reference direction determined in S11 and the actual working condition of the slide, the standardized dynamic boundary conditions for adaptive modeling are obtained by applying triaxial motion decoupling constraints, linear dynamic constraints, fixed sampling period discrete constraints and rigid connection constraints between the slide and the camera. S13. Based on the standardized dynamic boundary conditions obtained in S12, a unified parameter system required to construct the equations of motion is obtained by defining a parameter set including system input, mechanical parameters, state vector and output vector. S14. Based on the unified parameter system obtained in S13 and the triaxial load dynamic equations established based on Newton's second law, and combined with the first-order Euler method to discretize the continuous domain dynamic equations, a discretized state equation is constructed, the formula of which is: ； in, For the first The system state vector for each control cycle; This is the state transition matrix; The input matrix; S15. Based on the discretized state equations obtained in S14, and taking the three-dimensional spatial displacement at the end of the sliding table as the system output, construct the output equation, which is: ； in, This is the output matrix; S16, based on the discretized state equation of S14 and the output equation of S15, through matrix invertibility, system stability and pixel-level error verification, an effective three-axis slide table motion model that meets engineering accuracy is obtained.

3. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 2, characterized in that: The parameter set in S13 specifically includes: the three-axis motor input vector. Motor thrust / torque constant matrix Three-axis equivalent mass matrix Viscous damping coefficient matrix Three-axis velocity vector Three-axis position vector System state vector and system output vector ; Wherein, the system state vector For containing three-axis velocity vectors With the three-axis position vector A 6×1 dimensional vector, the system output vector It is a 3×1 dimensional vector representing the three-dimensional spatial displacement of the end of the slide.

4. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 3, characterized in that: The specific steps in S1 to establish the relationship between 3D motion parameters and image blur kernel and pixel position offset include: S17. Calculate the spatial displacement of the slide during the exposure time based on the three-dimensional motion parameters, and convert the spatial displacement into pixel displacement by combining the camera calibration parameters. Construct a motion blur kernel for the two-dimensional pixel plane, using the following formula: ； in, For the first The motion blur kernel function corresponding to the frame image; To obscure the extent of nuclear diffusion, and , , For the spatial displacement of the sliding table, , These are the calibration coefficients from spatial displacement to pixel displacement; S18, Constructing the first based on motion fuzzy kernel A motion imaging model for frame images to clearly define the correlation between pixel position offset and pixel point. The imaging formula is: ； in, The grayscale value of the image; The actual lighting intensity of the scene; For the first The exposure time of the frame; The motion fuzzy kernel for the slide table; It is noise.

5. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 4, characterized in that: In S18, based on the fact that the slide table remains in motion during the acquisition process, the pixel points are determined according to the movement of the slide table. The set of location change observations is as follows: ； in, , For the first Pixels in a frame image Relative to the pixels in the first frame image Pixel displacement in the X and Y directions.

6. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 5, characterized in that: The specific steps of S2 include: S21. Calibrate the camera's response function using a standard grayscale plate experiment. And using the inverse function of the camera response function Calculate the first The scene lighting corresponding to the frame, the first Frame image is The calculation formula is: ； in, For the first Within a frame, the estimated scene physical radiosity value corresponding to the pixel position after displacement; For the first In a frame, the reference pixel coordinates The original estimated value of the scene's physical radiosity corresponding to the actual pixel position after displacement by the sliding table; S22. Based on the pixel displacement generated by the sliding table maintaining its motion state during the acquisition process, Taylor linearization is used to establish a first-order approximation relationship between the current pixel and the displaced pixel to achieve image alignment and normalization. The formula for the first-order approximation is: ； in, For the first In a frame, the reference pixel coordinates Approximate value of scene physical radiosity estimate corresponding to the pixel position after displacement by the sliding table; For the first In the frame, the reference pixel coordinates do not take into account the slide displacement. The corresponding scene physical radiosity estimation baseline value; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; For the first In a frame, the reference pixel coordinates The estimated physical radiometric value of the scene is in Partial derivatives in direction; S23. Based on the first-order approximation of S22, perform rearrangement and transformation, and combine with the original S21. The formula for calculating exposure normalization is obtained to calculate the first... The normalized radiance estimate for the scene illumination in a frame is given by the following formula: ； in, For the first In a frame, the reference pixel coordinates The final estimated value of the scene's physical radiometric intensity after exposure normalization correction.

7. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 6, characterized in that: The formula for the multi-dimensional fusion weights constructed in S3 is: ； in, For multi-dimensional weight fusion; , , , These are the weighting coefficients; For clarity weight; Brightness weight; Noise weights; For positional weights.

8. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 7, characterized in that: Sharpness weight in S3 The Laplacian operator is used to calculate the edge intensity of the image. The calculation formula is as follows: ； in, For the Laplace operator; For the first In the frame image, the coordinates are The original grayscale value corresponding to the pixel; This represents the total number of frames in the multi-frame observation image; Brightness weight The calculation formula is: ； in, , For a reasonable brightness range; Noise weight The signal-to-noise ratio (SNR) of the image is determined by calculation, using the following formula: ； in, For the first Signal-to-noise ratio of a frame image; Position weight The calculation formula is: ； in, For the first The distance between a frame and a target frame; the smaller the distance, the stronger the positional correlation, and the greater the weight. .

9. The sequential image fusion and enhancement method based on triaxial slide displacement sensing according to claim 8, characterized in that: The specific steps of S4 include: S41. The normalized result obtained in S23 And S3 A weighted summation model is used to achieve multi-frame fusion, resulting in the optimal illumination estimate for the target frame. The formula is: ； S42, via camera response function S41 Map back to grayscale to obtain a high-quality single-frame fused image. The calculation formula is: 。

Images