A vortex dynamics image detail enhancement method for underground coal mine vision
By modeling image patch matching as fluid particle dynamics evolution and combining flow field advection migration and vortex core spiral capture, the problem of local optimum traps in complex texture regions of existing methods is solved, achieving high-precision detail enhancement and robust enhancement, which is suitable for underground coal mine vision applications.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA UNIV OF MINING & TECH
- Filing Date
- 2026-04-17
- Publication Date
- 2026-06-05
AI Technical Summary
Existing image detail enhancement methods are prone to getting stuck in local optima when dealing with complex textures or high-frequency details, leading to detail degradation and reduced structural consistency. Furthermore, they rely on large-scale computation or training with large amounts of data, making it difficult to meet real-time processing and generalization requirements.
Image patch matching is modeled as the dynamic evolution of fluid particles in a two-dimensional residual latent space. By constructing a dynamic switching mechanism between large-scale advection migration of the flow field and centripetal spiral capture of the vortex core, global search and local precise locking are achieved. A self-supervised residual learning framework is adopted to avoid greedy search.
It achieves high-precision feature extraction and enhancement in complex texture regions, suppresses artifacts and color distortion, has strong robustness and generalization ability, is suitable for real-time environments, and avoids the local optima trap and high training cost of traditional methods.
Smart Images
Figure CN122155978A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of digital image processing and computer vision technology, specifically relating to a method for enhancing the details of vortex dynamics images for underground coal mine vision. Background Technology
[0002] In recent years, with the popularization of digital imaging technology, images, as a key carrier of information transmission, directly affect the execution efficiency of downstream computer vision tasks such as target recognition, medical image analysis, and remote sensing monitoring. However, limited by the performance of optical lenses and the influence of complex environmental noise (such as insufficient lighting and sensor interference), acquired images often suffer from phenomena such as blurred details, texture degradation, and edge loss. Image detail enhancement techniques decompose images into a low-frequency smoothing layer containing the main structure and a detail layer containing high-frequency information, selectively amplifying the detail layer while maintaining the stability of the original information entropy, thereby enhancing texture expressiveness. Although existing technologies can improve image clarity to some extent, they still face significant challenges in balancing enhancement effects, execution efficiency, generalization ability, and model interpretability.
[0003] Existing image detail enhancement methods are mainly classified into four categories: local filtering-based, global optimization-based, deep learning-based, and residual learning-based methods. However, these methods often have limitations when processing complex textures or high-frequency details. Local filtering-based methods (such as bilateral filtering (BF) and guided filtering (GIF)) primarily rely on weighted smoothing based on the spatial distance and brightness similarity between pixel neighbors. While these methods are computationally efficient, they are limited by the perceptual range of the local window, making them prone to gradient inversion or artifacts (Halo phenomenon) when processing high-contrast edge regions. Furthermore, they are highly sensitive to parameters, making it difficult to balance accuracy and robustness in detail extraction under complex noise environments.
[0004] Global optimization-based methods (such as Weighted Least Squares (WLS)) Gradient minimization, etc.: Separation of smoothing and detail layers is achieved by constructing a global optimization objective function. Although this effectively alleviates edge halo effects, such models typically involve iterative solutions to large-scale sparse linear equations. Their computational complexity and memory consumption increase significantly with image resolution, making it difficult to meet the engineering requirements of high-resolution images or real-time processing. Furthermore, they are prone to causing the loss of local high-frequency details when processing complex texture regions.
[0005] Deep learning-based methods (such as DIP, CSGISNet, and DeepFSPIS) utilize the multi-layered nonlinear modeling capabilities of neural networks for feature learning. However, the performance of these methods is highly dependent on the size and labeling quality of the training dataset, and their generalization ability is often limited when dealing with unlabeled real-world scenarios or cross-domain tasks. Furthermore, the internal feature evolution process of deep neural networks lacks rigorous physical support, resulting in weak model interpretability, and the high training costs and hardware resource dependencies limit their widespread application in real-time embedded environments.
[0006] Traditional residual learning methods (such as zero-order inverse filtering (ZF) and in-situ residual homogeneity (IPRH)) aim to accurately reconstruct detail layers by capturing nonlinear components in the latent feature space. However, traditional residual learning methods rely excessively on greedy search mechanisms for local matching. Because their exploration scale is limited to the "in-situ" local neighborhood, the algorithms are prone to getting trapped in local optima when dealing with complex scenes with repetitive structures or drastic texture changes. They fail to fully explore and utilize potential globally optimal structural features, ultimately resulting in detail degradation and reduced structural consistency. Summary of the Invention
[0007] The purpose of this invention is to provide a method for enhancing the details of vortex dynamics images for underground coal mine vision. This method can overcome the limitations of traditional greedy search, take into account both global optimal search and local precise locking, and achieve accurate extraction and high-fidelity enhancement of detailed features while maintaining the stability of the overall image information structure.
[0008] To achieve the above objectives, the present invention provides a method for enhancing the details of vortex dynamics images for underground coal mine vision, comprising the following steps: S1, Image multi-scale decomposition and initial residual space construction; S2, Construct the composite fitness function and initialize the vortex particle system; S3, the early stage of system evolution, enters the large-scale global search stage based on the advection migration of the flow field; S4, the system evolution enters the middle and late stages, switching to the local high-precision development stage based on vortex core centripetal spiral capture; S5, Boundary Constraints, Steady-State Updates, and Final Image Reconstruction.
[0009] As a further aspect of the present invention: Step S1 specifically includes: The input low-resolution or degraded original image is converted into double-precision floating-point data, and each color channel is extracted. Each single-channel image is denoted as the original image. First, upsampling is performed using bilinear interpolation to obtain a high-resolution feature layer. ,Right now ,in, For upsampling operators; then... The original size is restored again by bilinear downsampling to obtain the basic smoothing layer. ,Right now ,in, The downsampling operator subtracts the original image from the base smoothing layer to obtain an initial residual containing high-frequency details. and will It serves as the starting point for the two-dimensional search space of subsequent dynamic evolution.
[0010] As a further aspect of the present invention: Step S2 specifically includes: setting in the two-dimensional residual space... A discrete fluid particle, the position vector of each particle. This represents the block matching coordinate offset on the feature map; pixel differences are calculated, and gradient edge and texture feature differences are introduced to construct a composite fitness function, defining the first... During the nth iteration fitness of a single particle for: ; in, For pixel absolute error, To compute the marginal differences using the first-order gradient operator, For high-frequency texture differences calculated based on standard deviation filtering, These are the corresponding weighting coefficients; Introducing the number of iterations Linearly decreasing control frequency This guides the system to transition from global exploration to local development. ; in, The initial frequency, This represents the maximum number of iterations. Based on control frequency Calculate the circulation intensity of a particle. To distinguish whether a particle is in a steady state or an unsteady state: ; in, and Representing respectively in the The minimum and maximum potential energy responses in the entire search group during round iteration. To avoid extremely small constants with a denominator of 0.
[0011] As a further aspect of the present invention: Step S3 specifically includes: Set behavior conversion threshold ,when At this point, the system is determined to be in the early stages of evolution. The system simulates the incompressibility of the fluid and generates a spatially mutually exclusive vector by calculating the local particle density within the induced radius of each particle. This causes particles to migrate to low-density regions; First, a local pressure intensity coefficient driven by the coupling of global control frequency and local density is defined. : ; in, Let be the total number of discrete particles in the system. Indicates the current iteration At that time, it fell into the first The radius of perception of a single particle The number of neighboring particles within the range is used to quantify the local spatial density in real time; Based on local pressure intensity coefficient Spatial mutually exclusive vectors The direction is defined as the direction of pressure release away from neighboring particles, expressed by the formula: ; In the formula, Represents the perception radius The set of neighborhood particles within; and These are the position vectors of the current particle and its neighboring particles, respectively. To prevent extremely small constants with a denominator of zero; During the advection migration phase, the logarithmic ratio of the average fitness of all particles in the current system to the optimal energy is calculated, and the advection induced vector is generated by combining the circulation intensity. Specifically, it is expressed as: ; in, In the first The optimal position in the round of iteration, In the first The average potential energy response of the entire particle swarm during round iteration. In the first In the first iteration The position of the search particle; Advection induced vector Guide particles to move in a wide-range advection towards the globally dominant vortex core; particle integrated spatial mutual exclusion vectors With advection induced vector The total displacement is obtained and its spatial coordinates are updated. .
[0012] As a further aspect of the present invention: Step S4 specifically includes: when At this time, the system enters the spiral capture phase, and the precession angle is... With circulation intensity Nonlinear coupling is employed to achieve adaptive adjustment of the search density within the optimal region; At the same time, the spiral radius by With vortex-induced scalar The constraints ensure that the search range smoothly converges towards the center during the iteration process, realizing the transition from wide-area exploration to neighborhood pixel locking. The relevant geometric relationship can be represented as: ; ; ; ; and The orthogonal components of the helical displacement vector in the two-dimensional search plane are represented by the resultant vector. This set of equations drives the particles to construct an evolutionary trajectory that balances the breadth of the global search with the accuracy of local convergence. Introducing the Lamb-Oseen vortex model to construct centripetal viscous damping Specifically, it is expressed as: ; in, It is the equivalent viscosity constant. This refers to the current iteration round; The particle adaptively adjusts the spiral rotation angle according to its own circulation strength. and combined with control frequency Given a helix radius, the particle approximates the global optimum along a logarithmic helix trajectory. The particle's position is precisely updated by calculating the product of the helix's orthogonal components and the distance vector, and then superimposing viscous damping and dynamic weights. Specifically, this is expressed as follows: .
[0013] As a further aspect of the present invention: Step S5 specifically includes: If the particle coordinates exceed the boundary of the set nonlocal search window during the evolution process... If the particle's motion is not properly controlled, an elastic reflection mechanism is used to remap it back into the effective space, preserving the particle's kinetic energy. Subsequently, the system calculates the fitness of the new position. A particle updates its physical coordinates only when the fitness of the new position is less than that of the current position, and updates the global optimal center after traversing all particles. ;go through After the evolution reaches a stable state, the residual blocks corresponding to the optimal offset coordinates are extracted, aggregated, and averaged to obtain the optimized precise detail layer. Finally, an enhancement factor is introduced. , and the original image Linear overlay is performed to output a high-resolution image with enhanced details. .
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention innovatively models the image patch matching and residual optimization process as the dynamic evolution of fluid particles in a non-uniform potential energy field. By constructing a dynamic switching mechanism between large-scale advection migration and centripetal spiral capture of the vortex core, it successfully establishes an optimal dynamic balance between global search breadth and local locking accuracy. This physical modeling approach fundamentally solves the technical problem of traditional image detail enhancement algorithms easily getting trapped in local optima in complex texture regions, leading to detail degradation and reduced structural consistency.
[0015] This invention overcomes the limitations of traditional greedy search mechanisms, achieving high-precision feature optimization across the entire global scope. Existing image detail enhancement methods based on residual learning rely on local greedy search strategies, which are prone to getting trapped in local optima when dealing with scenes with repetitive structures or drastic texture changes. This invention introduces the vortex evolution law from fluid dynamics, utilizing the background flow field for large-scale advection movement in the early stages of evolution to achieve global exploration, and relying on the strong centripetal attraction of the vortex core to drive particles along a logarithmic spiral trajectory to approach the optimal solution in the later stages of evolution. This adaptive optimization path, from broad to deep, effectively overcomes the blindness of the search and significantly improves the accuracy of locking onto complex high-frequency features.
[0016] This invention introduces multi-feature collaborative constraints, which significantly enhances high-frequency details while effectively suppressing artifacts and color distortion. Traditional local filtering (such as guided filtering in GIFs) and global optimization algorithms often suffer from severe visual defects such as edge halos (Halo phenomenon), gradient inversion, or color oversaturation when enhancing details. This invention constructs a composite fitness function that integrates pixel differences, gradient edges, and texture features, replacing a single pixel fidelity metric. It effectively suppresses the introduction of artificial artifacts and structural noise, maintaining the structural consistency and color fidelity of the original image.
[0017] This invention does not require pre-training on a large dataset and possesses generalization ability and strong robustness. Compared to existing deep learning detail enhancement models (such as DeepFSPIS), this invention adopts a self-supervised residual learning framework, which does not rely on large datasets or high hardware training costs. Attached Figure Description
[0018] Figure 1This is the overall framework diagram of the present invention; Residual Decomposition Stage Vortex Dynamics-inspired Optimization Stage Final Enhancement Stage Original Low-Quality Image Initial Residual Bilinear Upsampling Bilinear Downsampling Smooth Base Layer — Basic Smoothing Layer InitialResidual Patch — Initial Residual Patch Optimized Patches Current Patch Pending Patches Search Space Boundary Weak Vortex Strong Vortex Advection Streamlines Vector Synthesis of a Single Particle Decay of control parameter over iterations — control parameters decay iteration process Logarithmic Spiral Paths Optimized Residual Patch Optimized Residual (Detail Layer) — Optimizes the residual (detail layer). Enhanced High-Detail Image Gain Factor Figure 2 This is a comparison chart of objective metrics between this invention and other algorithms (all results are calculated using the Y channel. Best performance is indicated in bold, and second-best performance is indicated by underline). Figure 3 These are comparison images of the subjective visual effects of this invention in multiple scenarios; Figure 4 This is a comparison diagram of the error thermodynamics and spectral characteristics of the present invention. Detailed Implementation
[0019] A vortex dynamics-based image detail enhancement method for underground coal mine vision breaks through the limitations of traditional greedy search by modeling the image patch matching process as the dynamic evolution behavior of fluid particles in a two-dimensional residual latent space, such as... Figure 1 As shown, the overall architecture consists of three stages: initial residual decomposition, vortex dynamics heuristic optimization, and final image reconstruction. The specific steps are as follows: S1, Image multi-scale decomposition and initial residual space construction; Specifically, the input low-resolution or degraded original image is converted into double-precision floating-point data, and each color channel is extracted. Each single-channel image is denoted as the original image. To preserve the edge features of the image when extracting the residuals, this invention uses a linear operator to obtain the initial residuals, and first uses bilinear interpolation for upsampling to obtain a high-resolution feature layer. ,Right now ,in, For upsampling operators; then... The original size is restored again by bilinear downsampling to obtain the basic smoothing layer. ,Right now ,in, The downsampling operator subtracts the original image from the base smoothing layer to obtain an initial residual containing high-frequency details. The initial residual It can better map the true gradient field distribution of the image, and... It serves as the starting point for the two-dimensional search space of subsequent dynamic evolution.
[0020] S2, after constructing the initial residual space, construct the composite fitness function and initialize the vortex particle system; Specifically, in the two-dimensional residual space, we set... A discrete fluid particle, the position vector of each particle. This represents the block matching coordinate offset on the feature map; To accurately measure the fidelity between matching blocks, not only pixel differences are calculated, but also gradient edge and texture feature differences are introduced. Specifically, edge differences calculated by the first-order gradient operator and high-frequency texture differences calculated based on standard deviation filtering are used. These three difference metrics, combined with preset weighting coefficients, construct a composite fitness function, defining the first... During the nth iteration fitness of a single particle for: ; in, For pixel absolute error, To compute the marginal differences using the first-order gradient operator, For high-frequency texture differences calculated based on standard deviation filtering, These are the corresponding weighting coefficients; Introducing the number of iterations Linearly decreasing control frequency To guide the system's transition from global exploration to local development, the initial frequency is set to 2 to facilitate a smooth transition of system states. ; in, The initial frequency, This represents the maximum number of iterations. Based on control frequency Calculate the circulation intensity of each mass based on its current fitness value. To distinguish whether a particle is in a steady state or an unsteady state: ; in, and Representing respectively in the The minimum (optimal) and maximum (worst) potential energy responses in the entire search group during round iteration. To avoid extremely small constants with a denominator of 0.
[0021] S3, the early stage of system evolution, enters the large-scale global search stage based on the advection migration of the flow field; Specifically, set behavior conversion thresholds. As iterations progress, when At this point, the system determines that the evolution is in its early stages. To prevent premature aggregation of particles, the system simulates the incompressibility of the fluid and generates a spatially mutually exclusive vector by calculating the local particle density within the inductive radius of each particle. This encourages particles to migrate to low-density areas, thus maintaining population diversity.
[0022] Specifically, a local pressure intensity coefficient driven by the coupling of global control frequency and local density is first defined. : ; In the formula, Let be the total number of discrete particles in the system. Indicates the current iteration At that time, it fell into the first The radius of perception of a single particle The number of neighboring particles within the range is used to quantify the local spatial density in real time.
[0023] Based on local pressure intensity coefficient Spatial mutually exclusive vectors The direction is defined as the direction of pressure release away from neighboring particles, and its complete calculation formula is: ; In the formula, Represents the perception radius The set of neighborhood particles within; and These are the position vectors of the current particle and its neighboring particles, respectively. To prevent extremely small constants with a denominator of zero.
[0024] During the advection migration phase, in order to reasonably quantify the attraction intensity of the global optimal position to different search particles, the average fitness of all particles in the current system is calculated as the logarithm ratio of the optimal energy, and the advection induced vector is generated by combining the circulation intensity. Specifically, it is expressed as: ; in, In the first The optimal position in the round of iteration, In the first The average potential energy response of the entire particle swarm during round iteration. In the first In the first iteration The position of the search particle; Advection induced vector Guide particles to move in a wide-range advection towards the globally dominant vortex core; particle integrated spatial mutual exclusion vectors With advection induced vector The total displacement is obtained and its spatial coordinates are updated. This allows us to effectively escape the trap of local optima and complete a global, broad exploration of the solution space.
[0025] S4, the system evolution enters the middle and late stages, switching to the local high-precision development stage based on vortex core centripetal spiral capture; when At this point, the system enters the spiral capture phase. From a dynamic perspective, this process simulates the centripetal convergence behavior of particles near the dominant vortex core. Under the influence of centripetal force, the particles gradually approach the globally optimal solution along the spiral trajectory. At the trajectory modeling level, the potential field minimum indicates that the helical motion path of the particle possesses spatial adaptive characteristics. This is achieved by minimizing the precession angle. With circulation intensity Nonlinear coupling is employed to achieve adaptive adjustment of the search density within the optimal region.
[0026] At the same time, the spiral radius by With vortex-induced scalar The constraints ensure that the search range smoothly converges towards the center during the iteration process, realizing the transition from wide-area exploration to neighborhood pixel locking. The relevant geometric relationship can be represented as: ; ; ; ; and This represents the orthogonal components of the helical displacement vector in the two-dimensional search plane. This is achieved by defining the resultant vector. This set of equations drives the particles to construct an evolutionary trajectory that balances the breadth of the global search with the accuracy of local convergence.
[0027] To suppress overshoot during the local development phase, a Lamb-Oseen vortex model is introduced to construct centripetal viscous damping. Specifically, it is expressed as: ;
[0028] in, It is the equivalent viscosity constant. This refers to the current iteration round; The particle adaptively adjusts the spiral rotation angle according to its own circulation strength. and combined with control frequency Given a helix radius, the particle approximates the global optimum along a logarithmic helix trajectory. The particle's position is precisely updated by calculating the product of the helix's orthogonal components and the distance vector, and then superimposing viscous damping and dynamic weights. Specifically, this is expressed as follows: ; The spiral centripetal mechanism enables high-fidelity locking and fine-tuning of local weak textures and high-frequency edge features.
[0029] S5, Boundary Constraints, Steady-State Updates and Final Image Reconstruction; Specifically, after each dynamic position update, if the particle coordinates exceed the set nonlocal search window boundary during the evolution process... The system employs an elastic reflection mechanism to handle potential boundary overflow issues. It uses boundary values for mirror reflection calculations to remap the particles back into the effective space, preserving their kinetic energy. Subsequently, the system calculates the fitness of the new position. A particle updates its physical coordinates only when the fitness of the new position is less than that of the current position, and updates the global optimal center after traversing all particles. ;go through After the evolution reaches a stable state, the corresponding residual blocks are extracted based on the optimal offset coordinates obtained from each channel, and then aggregated and averaged to generate the optimized precise detail layer. Finally, an enhancement factor is introduced. , and the original image Linear overlay is performed, and after truncation of pixel values (upper and lower limits), a high-resolution image with enhanced detail is output. .
[0030] like Figure 2 The quantitative comparison results show that, in tests with different magnifications on multiple benchmark datasets (such as RealSRSet, BSDS200, and T91), the method proposed in this invention achieves optimal performance in objective evaluation metrics such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), surpassing current mainstream global optimization algorithms and traditional residual learning methods. Furthermore, combined with… Figure 3 The comparison of subjective visual enhancement effects shows that the enhancement results of this invention effectively avoid edge overshoot, ghosting (staircase effect), and color oversaturation phenomena common in other algorithms. The image presents a more natural and realistic visual effect, and the fitting degree of complex textures and geometric edges is highly consistent with the real distribution. Furthermore, as... Figure 4 As shown in the error heatmap and spectral distribution characteristic analysis, the present invention exhibits extremely low pixel error residue in the global space (uniform dark blue in the heatmap), and the enhanced high-frequency spectral energy distribution perfectly matches the physical laws of the real image, without introducing additional structural noise or frequency distortion. This fully demonstrates the superior performance of the present invention in high-fidelity restoration of details in single images, possessing significant engineering application value and broad prospects for practical deployment.
Claims
1. A method for enhancing the details of vortex dynamics images for underground coal mine vision, characterized in that, Includes the following steps: S1, Image multi-scale decomposition and initial residual space construction; S2, Construct the composite fitness function and initialize the vortex particle system; S3, the early stage of system evolution, enters the large-scale global search stage based on the advection migration of the flow field; S4, the system evolution enters the middle and late stages, switching to the local high-precision development stage based on vortex core centripetal spiral capture; S5, Boundary Constraints, Steady-State Updates, and Final Image Reconstruction.
2. The method for enhancing the detail of vortex dynamics images for underground coal mine vision according to claim 1, characterized in that, The specific steps in S1 include: The input low-resolution or degraded original image is converted into double-precision floating-point data, and each color channel is extracted. Each single-channel image is denoted as the original image. First, upsampling is performed using bilinear interpolation to obtain a high-resolution feature layer. ,Right now ,in, For upsampling operators; then... The original size is restored again by bilinear downsampling to obtain the basic smoothing layer. ,Right now ,in, The downsampling operator subtracts the original image from the base smoothing layer to obtain an initial residual containing high-frequency details. and will It serves as the starting point for the two-dimensional search space of subsequent dynamic evolution.
3. The method for enhancing the detail of vortex dynamics images for underground coal mine vision according to claim 2, characterized in that, The specific steps of S2 include: setting in the two-dimensional residual space... A discrete fluid particle, the position vector of each particle. This represents the block matching coordinate offset on the feature map; pixel differences are calculated, and gradient edge and texture feature differences are introduced to construct a composite fitness function, defining the first... During the nth iteration fitness of a single particle for: ； in, For pixel absolute error, To compute the marginal differences using the first-order gradient operator, For high-frequency texture differences calculated based on standard deviation filtering, These are the corresponding weighting coefficients; Introducing the number of iterations Linearly decreasing control frequency This guides the system to transition from global exploration to local development. ； in, The initial frequency, This represents the maximum number of iterations. Based on control frequency Calculate the circulation intensity of a particle. To distinguish whether a particle is in a steady state or an unsteady state: ； in, and Representing respectively in the The minimum and maximum potential energy responses in the entire search group during round iteration. To avoid extremely small constants with a denominator of 0.
4. The method for enhancing the detail of vortex dynamics images for underground coal mine vision according to claim 3, characterized in that, The specific steps of S3 include: Set behavior conversion threshold ,when At this point, the system is determined to be in the early stages of evolution. The system simulates the incompressibility of the fluid and generates a spatially mutually exclusive vector by calculating the local particle density within the induced radius of each particle. This causes particles to migrate to low-density regions; First, a local pressure intensity coefficient driven by the coupling of global control frequency and local density is defined. : ； in, Let be the total number of discrete particles in the system. Indicates the current iteration At that time, it fell into the first The radius of perception of a single particle The number of neighboring particles within the range is used to quantify the local spatial density in real time; Based on local pressure intensity coefficient Spatial mutually exclusive vectors The direction is defined as the direction of pressure release away from neighboring particles, expressed by the formula: ； In the formula, Represents the perception radius The set of neighborhood particles within; and These are the position vectors of the current particle and its neighboring particles, respectively. To prevent extremely small constants with a denominator of zero; During the advection migration phase, the logarithmic ratio of the average fitness of all particles in the current system to the optimal energy is calculated, and the advection induced vector is generated by combining the circulation intensity. Specifically, it is expressed as: ； in, In the first The optimal position in the round of iteration, In the first The average potential energy response of the entire particle swarm during round iteration. In the first In the first iteration The position of the search particle; Advection induced vector Guide particles to move in a wide-range advection towards the globally dominant vortex core; particle integrated spatial mutual exclusion vectors With advection induced vector The total displacement is obtained and its spatial coordinates are updated. .
5. A method for enhancing the detail of vortex dynamics images for underground coal mine vision according to claim 4, characterized in that, The specific steps of S4 include: when At this time, the system enters the spiral capture phase, and the precession angle is... With circulation intensity Nonlinear coupling is employed to achieve adaptive adjustment of the search density within the optimal region; At the same time, the spiral radius by With vortex-induced scalar The constraints ensure that the search range smoothly converges towards the center during the iteration process, realizing the transition from wide-area exploration to neighborhood pixel locking. The relevant geometric relationship can be represented as: ； ； ； ； and The orthogonal components of the helical displacement vector in the two-dimensional search plane are represented by the resultant vector. This set of equations drives the particles to construct an evolutionary trajectory that balances the breadth of the global search with the accuracy of local convergence. Introducing the Lamb-Oseen vortex model to construct centripetal viscous damping Specifically, it is expressed as ； in, It is the equivalent viscosity constant. This refers to the current iteration round; The particle adaptively adjusts the spiral rotation angle according to its own circulation strength. and combined with control frequency Given a helix radius, the particle approximates the global optimum along a logarithmic helix trajectory. The particle's position is precisely updated by calculating the product of the helix's orthogonal components and the distance vector, and by superimposing viscous damping and dynamic weights. Specifically, this is expressed as follows: .
6. A method for enhancing the detail of vortex dynamics images for underground coal mine vision according to claim 5, characterized in that, The specific steps of S5 include: If the particle coordinates exceed the boundary of the set nonlocal search window during the evolution process... If the particle's motion is not properly controlled, an elastic reflection mechanism is used to remap it back into the effective space, preserving the particle's kinetic energy. Subsequently, the system calculates the fitness of the new position. A particle updates its physical coordinates only when the fitness of the new position is less than that of the current position, and updates the global optimal center after traversing all particles. ;go through After the evolution reaches a stable state, the residual blocks corresponding to the optimal offset coordinates are extracted, aggregated, and averaged to obtain the optimized precise detail layer. Finally, an enhancement factor is introduced. , and the original image Linear overlay is performed to output a high-resolution image with enhanced details. .