Mine image detail enhancement method based on fish eagle optimization algorithm

Abstract

The application relates to the technical field of image enhancement processing, in particular to a mine image detail enhancement method based on a fish eagle optimization algorithm, which comprises the following steps: constructing an image detail enhancement model based on residual learning and redesigning an energy function Y(z); extracting an initial residual part through the difference between a scaled image and an original image; using a fish eagle optimization algorithm to improve a matching method to find more suitable detail layer features, so that the system is easier to converge to a global optimal solution; manually adjusting an amplification factor parameter to amplify the obtained detail layer, and superimposing the amplified detail layer with an input image I0, so that an image I after detail enhancement is obtained e ; calculating enhanced image pixel values from the reconstructed image gradient, the original image and the gradient thereof, and finally reconstructing the image. The application optimizes residual features by designing a feature refinement mechanism based on the fish eagle optimization algorithm, so that the image processing has strong global search capability, can effectively avoid falling into a local optimal solution, and improves the image detail processing quality.

Description

Technical Field

[0001] This application relates to the field of image enhancement processing technology, and in particular to a method for enhancing the details of mine images based on the Osprey optimization algorithm. Background Technology

[0002] In the mining environment, insufficient lighting, noise interference, and pervasive dust are common problems, resulting in low contrast and blurred details in acquired images. This poor image quality not only affects the intuitive identification of the work area but also hinders critical tasks such as crack detection and equipment condition monitoring. Furthermore, mine images often require the extraction of subtle features from low-quality data; missing details may lead to undetected hazards, directly impacting the safety and efficiency of mine operations. Therefore, enhancing the detail, clarity, and recognizability of mine images has become a core issue in image processing for mining environments.

[0003] Image detail enhancement is a key technology in computer vision with broad application prospects. It is not only widely integrated into consumer electronics and medical imaging devices, such as smartphones, televisions, tablets, and endoscopes, but also crucial for improving image quality. Detail enhancement algorithms are particularly valuable for low-resolution or poor-quality images, especially in the mining industry, as they help workers better understand the complex environment inside mines, enabling more accurate decision-making and improving mining production efficiency and safety.

[0004] Currently, in mine image processing, when using search matching methods to extract and refine residual features, the focus is usually only on pixel loss, neglecting the effective processing of image edges and texture structures. Because images in mine environments inherently suffer from low illumination and high noise, this singular optimization objective easily leads to a decline in the quality of the enhanced image, manifesting as blurred edges, distorted textures, or unnatural artifacts, thus affecting the identification of key details and subsequent analysis.

[0005] Residual learning-based algorithms typically employ search-matching techniques to extract detailed layers from images. However, these search processes are mostly based on greedy mechanisms, which can easily lead to premature local optima in the complex image data of mining environments. Furthermore, they fail to fully explore the entire search space and converge to a local optimum with a high probability, thus affecting the overall effect and performance of detail enhancement and posing challenges to the visualization and monitoring of mining operations and the extraction of key features. Summary of the Invention

[0006] The technical problem this invention aims to solve is that existing image enhancement algorithms have poor recognition performance in image processing in mining environments, and tend to get trapped in local optima too early during detail enhancement, resulting in poor overall detail enhancement performance.

[0007] Therefore, this invention provides a method for enhancing the details of mine images based on the Osprey optimization algorithm.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] A method for enhancing details in mine images based on the Osprey optimization algorithm includes the following steps:

[0010] Step 1: Construct an image detail enhancement model based on residual learning and redesign the energy function Y(z);

[0011] Step 2: Extract the initial residual by comparing the scaled image with the original image;

[0012] Step 3: Improve the matching method using the Osprey optimization algorithm to find more suitable detailed layer features, making it easier for the system to converge to the global optimum.

[0013] Step four: Manually adjust the magnification factor parameter to magnify the obtained detail layer, and then overlay it with the input image I0 to obtain the image I after detail enhancement. e .

[0014] Further, in step one, the detail enhancement algorithm based on residual learning is described as follows: Res=argmin∑ z∈ω |D(z+c(z))-D(z)|,I e = I + β × F(Res), where z is a region block, c(z) is the offset, D(z) is the region block centered at z in image I, ω is the feasible region of z, and I e Let represent the final detail-enhanced image, and β be the magnification factor of the detail layer. The initial residual features Res of D(z) are obtained by calculating all image region blocks.

[0015] Furthermore, the energy function Y(z) takes into account the pixel loss Y p (z) and two regularization terms Y g (z) and Y s (z), Y(z)=Y p (z)+γY g (z)+λY s (z), The energy function Y(z) is a new block matching criterion, which is optimized to obtain c(z), i.e., c(z) = argminY(z); and Let represent the Hamiltonian operator and the Laplacian operator for a certain region block D(z), respectively, where ω is the feasible region of z, and γ and λ represent the regularization constants that control the prior components.

[0016] Furthermore, in step three, a feature refinement mechanism F(~) is designed based on the improved matching method of the Osprey optimization algorithm to find more suitable detail layer features, further refine the initial residual features of Res, and finally obtain the detail layer F(Res), which converges to obtain the global optimal solution.

[0017] Furthermore, the improved matching method based on the Osprey optimization algorithm specifically includes:

[0018] S3.1 Random initialization of the osprey population

[0019] S3.2 By weighted summing of brightness, gradient, and texture features, the energy value of each pixel in I1 and the energy value of its corresponding Osprey agent in image I2 are calculated. Taking the absolute difference between these values ​​yields the fitness values ​​of the N Osprey agents corresponding to each pixel. After comparison Find the current best fitness value in I2.

[0020] S3.3 Find a reference location for each Osprey agent. This serves as the target for the next move, thus updating the position.

[0021] S3.4 Reference Position Update the position to which the next decision will be made;

[0022] S3.5 will update the Once the cropping reaches the boundary, a fitness-based selection mechanism is introduced to calculate the fitness value of the new position. Filter for the optimal position;

[0023] S3.6 will be the final position after iteration. Mapping onto the initial residual image H0, we calculate the weighted average value of each point, averageh0. We then remove the averageh0′ from the values ​​of the expanded boundary portion, and finally obtain the updated residual H1 = averageh0′.

[0024] Further, in step S3.2, the energy value of each pixel in I1 and the energy value at the corresponding position in I2 are subtracted, and the absolute value is taken to obtain the fitness value. Compare the N Osprey agents corresponding to each pixel in I1. Find the current best fitness value in I2. And let its position be considered the "best" position, that is, the set of pixel positions with the smallest fitness value, denoted as Where N is the number of ospreys in the population, and t is the number of iterations. N Osprey agents corresponding to each pixel The index of the minimum value individual. The coordinates of the indexed individual.

[0025] Furthermore, in step S3.3, for the p-th pixel in image I1, in the t-th iteration, the assigned N search agents search for a set of agents in image I2 that have a better fitness than themselves. By combining the selection of neighborhood solutions and global optimal solutions, the particle is guided to move in a more favorable direction, with a reference position. in In the set A randomly selected index in the middle. rand is a random number that follows a uniform distribution [0,1].

[0026] Furthermore, in step 3.4, during the t-th iteration, the i-th agent dispatched by the p-th pixel of image I1 is at the current position. Based on this, the location is updated. in, For the i-th agent at the current position, As a guiding point, Let c be the optimal position among the N agents, μ be the local guiding weight, I be the random factor, and rand(1,dim) be the random perturbation term, which is a randomly generated matrix of size 1×dim.

[0027] Furthermore, in step S3.5, the new position is... The fitness value of the new position is calculated by taking the absolute value of the difference between the energy value and the energy value of the corresponding pixel p in image I1. according to Select the current optimal value through comparison. Right now if The global optimum is then updated to The global optimal position is also updated accordingly. otherwise Then assign for for That is, at the start of the next iteration, the fitness and position of the first Osprey agent will be forcibly changed to... and Together with the other N-1 osprey agents, they form a new population of N.

[0028] Furthermore, in step 3.6, the cumulative matrix sumh0 and the counting matrix counterh0 are updated by cumulative stacking, and the weighted average value of each point is calculated.

[0029] The beneficial effects of this invention are that it leverages the Osprey Optimization Algorithm (OOA) to provide strong global search capabilities when processing images, effectively avoiding getting trapped in local optima. By simulating the exploration and development behavior of an osprey during predation, the algorithm searches for optimal solutions within a large search space. Simultaneously, random perturbations increase the diversity of the search, further improving convergence accuracy. Furthermore, the algorithm fully considers the high-frequency information of the image, preserving image details and edge characteristics through residual extraction and feature matching, thereby improving the visual quality of the processed image. In its specific implementation, the algorithm divides the image into multiple small blocks and measures similarity based on features such as brightness, gradient, and texture, ensuring that candidate regions can well reflect the feature information of the original image. The Osprey Optimization process dynamically adjusts the position of candidate solutions and iteratively updates them in conjunction with the objective function value, making the matching results more accurate. Finally, by averaging the high-frequency residuals, image details are effectively restored, making the algorithm perform excellently in detail enhancement tasks. Attached Figure Description

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0031] Figure 1 This is a flowchart of the mine image detail enhancement method based on the Osprey optimization algorithm in this invention.

[0032] Figure 2 This is a block diagram of the algorithm structure of the mine image detail enhancement method based on the Osprey optimization algorithm in this invention.

[0033] Figure 3 This is a schematic diagram comparing the image detail enhancement algorithm of this invention with other existing algorithms to process an image of a miner operating a coal mining machine with four times the detail magnification.

[0034] Figure 4 This is a schematic diagram comparing the image detail enhancement algorithm of this invention with other existing algorithms to process an image of a miner assessing ore conditions with a four-fold magnified detail. Detailed Implementation

[0035] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0036] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, features defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0037] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0038] A mine image detail enhancement method based on the Osprey optimization algorithm is proposed. This method addresses the issue that using search-matching techniques to obtain image detail layers often results in the search process converging to a local optimum, and that residual feature extraction and refinement only consider pixel loss while neglecting image edges and texture structure. The proposed algorithm leverages the Osprey optimization algorithm to find the global optimum, improving image detail enhancement performance with a slight sacrifice in processing speed. Specifically, firstly, a residual learning-based image detail enhancement model is proposed, and two new regularization terms force the matching process to consider not only pixel differences but also edge sharpness and texture smoothness, thus redesigning the energy function Y(z). Secondly, for extracting initial high-frequency details, the input image is scaled to obtain initial residual features. Next, the search-matching method is improved using the Osprey optimization concept, and the residual features are updated using the newly designed energy function Y(z) to obtain more suitable detail features. Finally, the magnification factor can be adjusted to enlarge the obtained detail layers, resulting in the final enhanced image through the image detail enhancement model. This invention proposes a more advanced search and matching technique based on the Osprey optimization algorithm, which obtains globally optimal detail features when updating residual features, thereby obtaining visually superior detail-enhanced images.

[0039] This application overcomes the problem of current algorithms easily getting trapped in local optima and neglecting detailed structures in mine image processing by designing a more efficient and comprehensive search and matching strategy and effectively integrating pixel loss and image structure information. This not only more accurately and comprehensively captures blurred edges and complex texture features in mine images, but also significantly improves the quality and naturalness of enhanced images, providing a superior solution for image processing in low-light, high-noise mining environments.

[0040] Specifically, the mine image detail enhancement method based on the Osprey optimization algorithm in this application includes the following steps:

[0041] Step 1: Construct an image detail enhancement model

[0042] We construct an image detail enhancement model based on residual learning and use two new regularization terms to force the matching process to pay attention not only to pixel differences, but also to edge sharpness and texture smoothness, thereby redesigning the energy function Y(z) to capture the local and global features of the image more comprehensively.

[0043] S1.1 Constructing an image detail enhancement model based on residual learning

[0044] The detail enhancement algorithm based on residual learning can be described by equations (1)-(3):

[0045]

[0046] Res=argmin∑ z∈ω |D(z+c(z))-D(z)| (2)

[0047] I e =I+β×F(Res) (3)

[0048] Where z is a region block, c(z) is the offset, D(z) is the region block centered at z in image I, ω is the feasible region of z, and I e Let β represent the final enhanced detail image, and β be the magnification factor of the detail layer. The initial residual feature Res, D(z), can be obtained by calculating all image region blocks. Then, due to the coarsness of the initial residual feature, a mechanism F(~) needs to be designed to update this feature. This invention further refines Res through search matching, ultimately obtaining the detail layer F(Res).

[0049] S1.2 Design Energy Function

[0050] The proposed minimization function Y(z) in this paper, compared to traditional search matching algorithms that only consider pixel loss Y, minimizes the energy function Y(z). p (z) Two new regularization terms Y have been added. g (z) and Ys (z). The purpose of this design is to consider not only pixel loss during the matching process, but also edge sharpness and texture smoothness.

[0051] Y(z)=Y p (z)+γY g (z)+λY s (z) (4)

[0052]

[0053] The energy function Y(z) is a new block matching criterion, which is optimized to obtain c(z), i.e., c(z) = argminY(z); and Let represent the Hamiltonian operator and the Laplacian operator for a certain region block D(z), respectively, where ω is the feasible region of z, and γ and λ represent the regularization constants that control the prior components.

[0054] The second step is to extract the initial residual by comparing the scaled image with the original image.

[0055] The initial residual, also known as the initial high-frequency detail, is the residual signal obtained through bilinear interpolation. It only has the same resolution as the original signal and cannot be considered as a detail layer to be superimposed; in short, it is an estimate of the detail layer signal. The core idea of ​​extracting the initial residual is based on multi-scale analysis. It decomposes the input image into low-frequency components and high-frequency residual information, using image scale changes and interpolation methods to simulate low-frequency smoothing, while simultaneously extracting high-frequency details through difference calculations.

[0056] First, the input image I0 is defined as the original image, upon which all subsequent analysis and processing are based. Next, bilinear interpolation is used to enlarge the original image I0 by a factor of 1.25, generating a larger image I1. This enlargement operation preserves most of the structural information of the original image while introducing a smoothing effect, making edge and texture features less noticeable.

[0057] To simulate the information loss caused by image scaling, I1 is further reduced back to its original size, resulting in image I2. Through this "scale-up then scale-down" process, I2 becomes a smoothed version of I0, preserving low-frequency information (such as large color blocks and shape contours) but losing many high-frequency details (such as edges, textures, and sharp changes). This operation is equivalent to low-pass filtering the image, producing a smooth approximation. The high-frequency information H0 is extracted by subtracting the original image I0 from the smoothed image I2. This high-frequency residual information represents the edges, textures, and rapidly changing details in the image. Mathematically, it can be considered an approximation of the high-frequency components in the image's frequency space, i.e., the initial residual image.

[0058] H0=I2-I0 (8)

[0059] The third step involves using the Osprey optimization algorithm to improve the matching method and find more suitable details layer features. This further refines the initial residual features of Res, ultimately obtaining the details layer F(Res), making it easier for the system to converge to the global optimum.

[0060] In the process of finding the optimal matching block, almost all search methods, except for meaningless complete searches, are prone to getting trapped in local optima. This is because the search matching method is a greedy algorithm, which allows the energy function to only accept states better than the current state during iteration. This makes the system easily trapped in local optima and unable to find the global optimum. To find the global optimum, this invention designs a new feature refinement mechanism F(~), which is based on the residual matching mechanism of the Osprey optimization algorithm.

[0061] The novel feature refinement mechanism F(~) is designed based on the Osprey optimization algorithm, an intelligent optimization algorithm that simulates natural hunting behavior and has good global search and local exploitation capabilities. By dynamically updating position, random perturbation, a competition mechanism, and multi-feature fitness evaluation, it effectively avoids local optima and quickly converges to the global optimum. The specific design method is as follows:

[0062] The proposed object-oriented analysis is a population optimization method that seeks suitable solutions by utilizing the exploration and development capabilities of population members in the problem-solving space through an iterative search process. In the Osprey Optimization (OOA) algorithm, each osprey is considered an exploration agent, and its position represents a candidate solution to the problem variable. The entire population consists of multiple ospreys and is modeled using mathematical vectors. In the initial stage of the OOA algorithm, the positions of the ospreys in the search space are randomly initialized using formula (9), thereby ensuring good diversity and exploration capabilities in the search process.

[0063] S3.1 First, the osprey population is randomly initialized:

[0064]

[0065] i is the index of an individual osprey in the population, and N is the total number of ospreys in the population. In this invention, it refers to the number of exploration agents, that is, the number of candidate search blocks used for feature matching and optimization in the image region. t is the current iteration number, and p is the p-th pixel in the I1 image. If the size of I1 is M×N, then the range of p is [1,MN]. Therefore, the above formula expresses the exploration agent dispatched by the p-th pixel in I1 in the current iteration round t. dim represents the dimension of the osprey's position, which should be 2 in this invention, that is, the position of the x-coordinate or y-coordinate of the i-th agent. rand(N,dim) represents a randomly generated matrix of size N×dim, where each element is a random number uniformly distributed in the interval [0,1]. LB and UB represent the lower and upper bounds of the search space, respectively, to ensure that the exploration range does not exceed the image boundary.

[0066] S3.2 Then, by using equation (4) to calculate the energy value of each pixel in I1 by weighted summation of brightness, gradient and texture features, we can also calculate the energy value of the corresponding position of the Osprey agent dispatched by them in image I2. The fitness value can be obtained by subtracting the energy value of each pixel from the energy value of the corresponding position and taking the absolute value. By comparing the N Osprey agents corresponding to each pixel in I1 The best fitness value can be found in I2. And let its position be considered the "best" position, that is, the set of pixel positions with the smallest fitness value, denoted as .

[0067]

[0068] Based on comparing the N Osprey agents corresponding to each pixel Find the minimum value among N individuals in the t-th iteration of the p-th pixel, and extract the index of this minimum value individual. That is, the location coordinates of this indexed individual.

[0069] S3.3 then moves on to the exploration section, which is also the core part, and its purpose is to provide each Osprey agent with... Find a reference point This serves as the target for the next move, thus updating the position. First, for the p-th pixel in image I1, in the t-th iteration, the N search agents we assigned search for a set of agents in image I2 with better fitness than themselves. The specific design principle is shown in the following formula:

[0070]

[0071] In equation (12), j represents each individual i in the population. The index of an individual whose value is smaller than its own. Assume the total number of ospreys N in the population is 4, and their... The values ​​are 1, 2, 3, and 4. We take the first one, i = 1, and its value... His score is 1, compared to the other three. The smallest value is j, so j is 0; then we take the second value, which is when i = 2. Since the value is 2, the only value smaller than it is the first one, so j is 1, and so on.

[0072] We also use random conditions to determine whether to select the current global optimum as the target position. This design enhances the algorithm's global exploration capability, allowing the agent to escape local regions and thus expand the search range. Simultaneously, randomness introduces more uncertainty and diversity into the algorithm, preventing the search from getting stuck in fixed patterns or local optima. By combining the selection of neighborhood solutions and the global optimum, the algorithm achieves a balance between development and exploration, both deeply exploring the potential of local regions and guiding particles towards better directions using global information. The specific design is shown below:

[0073]

[0074] in In the set A randomly selected index in the middle. `rand` is a random number that follows a uniform distribution [0,1]. When no better fit is found in the neighborhood than the current agent, or when a random value is less than 0.5, a reference position is chosen. If a better agent exists in the neighborhood than the current Osprey and the random value is greater than or equal to 0.5, a better agent is randomly selected from the neighborhood. This serves as a reference point. It encourages local exploration, utilizing better solutions in the neighborhood to delve deeper into potential locally optimal regions. By randomly selecting neighborhood solutions, it enhances the diversity of the search and prevents all individuals in the population from concentrating on moving towards the same target.

[0075] After S3.4 is completed, the next decision-making position should be updated based on the previously obtained reference position. We can try updating the position using the following formula:

[0076]

[0077] In the t-th iteration, the i-th agent dispatched by the p-th pixel is at the current position. The strategy involves updating the location based on the current guide point. It also incorporates the optimal position among the current N proxies. The influence of the two guiding directions is modulated by the local guiding weight c and the global guiding weight μ, respectively. It's easy to see that the parameters c and μ gradually decrease with the number of iterations, thus strengthening exploration in the early stages of the algorithm and enhancing convergence in the later stages. The introduction of the random perturbation term rand(1,dim) in the update formula helps maintain the diversity of individual updates and avoids getting trapped in local optima. In addition, to enhance the dynamic change of the guiding direction, a random factor I is introduced, where I = round(1 + rand), and round represents rounding the result to the nearest integer. Therefore, the value of I may be 1 or 2. It can control with a certain probability whether the agent fully moves towards the target or partially adjusts its direction in each iteration, thereby further improving the stability and search capability of the algorithm.

[0078] S3.5 finally needs to be updated. Crop to the boundary area:

[0079]

[0080] new location The fitness value of the new location can be calculated by taking the absolute value of the difference between the energy value and the energy value of the corresponding pixel p in image I1. according to Select the current optimal value through comparison. Right now:

[0081]

[0082] if The global optimum is then updated to The global optimal position is also updated accordingly. otherwise Then assign for for That is, at the start of the next iteration, the fitness and position of the first Osprey agent will be forcibly changed to... and Together with the other N-1 Osprey agents, a new population of N is formed. This is the so-called elite preservation strategy. This idea is used to ensure that the optimal solution is continuously preserved in the optimization process, avoiding the loss of the optimal solution during iteration. By preserving the current best solution, convergence can be accelerated and degenerate to a worse solution due to the exploration of new solutions. It also helps to maintain the stability of the search process and increases the probability that the algorithm will find the global optimum.

[0083] S3.6 will be the final position after iteration. Mapping onto the initial residual image H0, and taking a 5×5 patch centered at each coordinate point, the cumulative matrix sumh0 and the counting matrix counterh0 are updated by cumulative stacking. The weighted average value for each point can be calculated using the following formula:

[0084]

[0085] Then, the value of the expanded boundary part is removed to obtain averageh'0, and the final update residual we obtain is H1 = averageh'0.

[0086] The fourth step involves manually adjusting the magnification factor parameters to magnify the obtained detail layer, overlaying it with the input image I0, and then updating the residuals to obtain the detail-enhanced image I. e This is used to ensure that better visual effects can be achieved in order to generate images with enhanced details.

[0087] The fifth step involves calculating the enhanced image pixel values ​​from the reconstructed image gradient, the original image, and its gradient, and finally reconstructing the image.

[0088] The overall process of obtaining the best matching image patch using Osprey optimization is shown in the pseudocode:

[0089] Algorithm pseudocode

[0090] Input: Population size N, upper and lower bounds of the exploration range UB and LB, boundary fill size parameter hfs, optimization parameter dimension dim, maximum number of iterations T, cmax, cmin, μmax, μmin.

[0091] The randomly initialized osprey population obtained through equation (9):

[0092] For t=1to T

[0093] For i = 1 to N

[0094] The global best fitness value and the corresponding optimal position are obtained through equations (10) and (11). and

[0095] The reference position of each Osprey agent is obtained through equation (13).

[0096] Randomness is introduced through equations (14) and (15) and... The process of updating candidate solution positions is guided by a method that avoids the Osprey going out of its exploration range and also avoids getting trapped in local optima.

[0097] End

[0098] According to equations (16) and (17) Select the current optimal value from the list. and its location index

[0099] Will and To update by comparison It will also be updated accordingly.

[0100] Using an elite preservation strategy, the position and fitness of the first osprey in the initial population of the next iteration are forcibly changed to... and It forms a new population of N individuals with the other N-1 individuals.

[0101] End

[0102] The location of the convergence point is obtained from the final iteration using equation (18). By calculating the weighted average of each residual point on the initial residual image H0, and then removing the boundary padding, the final refined detail layer H1 can be obtained.

[0103] Multiplying H1 by the amplification factor parameter and then superimposing it with I0 yields the final high-frequency detail reconstruction image I. e .

[0104] In summary, this invention utilizes the Osprey Optimization Algorithm (OOA) to provide strong global search capabilities when processing images, effectively avoiding getting trapped in local optima. By simulating the exploration and development behavior of an osprey during hunting, the algorithm searches for optimal solutions within a large search space. Furthermore, random perturbations increase the diversity of the search, further improving convergence accuracy. In addition, the algorithm fully considers the high-frequency information of the image, preserving image details and edge characteristics through residual extraction and feature matching, thereby improving the visual quality of the processed image.

[0105] In its implementation, the algorithm divides the image into multiple small patches and measures similarity based on features such as brightness, gradient, and texture to ensure that candidate regions can well reflect the feature information of the original image. The Osprey optimization process dynamically adjusts the positions of candidate solutions and iteratively updates them in conjunction with the objective function value, making the matching results more accurate. Finally, by averaging the high-frequency residuals, image details are effectively restored, making the algorithm perform excellently in detail enhancement tasks.

[0106] This method aims to improve the visual quality of mine images and addresses the problem of easily getting trapped in local optima when searching for the most similar image patch. Specifically, the input image I0 undergoes two bilinear interpolation operations to obtain image I2. The initial high-frequency residual image H0 is extracted by calculating the difference between the original image I0 and the downsampled image I2. In the residual optimization stage, the Osprey optimization algorithm is used to iteratively match and update the high-frequency residual image. First, this algorithm fully considers pixel loss during optimization while focusing on edge and texture structure features. By redesigning the energy function Y(z), it better preserves the local details and overall texture information of the image. Then, it performs matching optimization on H0, finding the optimal matching patch through local and global searches, generating an improved feature map H1. Finally, the optimized detail layer H1 is multiplied by a magnification factor and fused with the input image I0 to generate a detail-enhanced image I0. e .

[0107] The technical effects of this invention will be described in detail below, combining performance testing and experimental analysis. Existing image detail enhancement algorithms can be divided into the following categories: first, algorithms based on global filters, such as BFLS, TH, and ILS; second, algorithms based on local filters, including GGIF, EGIF, and RGIF; and third, algorithms based on residual learning, such as ZF and IPRH. These algorithms each have their advantages and disadvantages in detail enhancement tasks. This invention combines multiple optimization ideas to propose a superior enhancement method. To verify the algorithm's performance, this invention selected three internationally widely used natural image datasets: RealSRSet, BSDS200, and T91. These datasets each have their own characteristics: RealSRSet contains a large number of complex scenes, which can test the algorithm's performance on highly complex images; BSDS200 has a rich variety of image types, making it an ideal dataset for studying the generality of image processing; and T91 mainly consists of high-resolution images, which can effectively evaluate the algorithm's performance in high-resolution environments. Using these datasets for testing allows for a comprehensive measurement of the algorithm's overall performance under different scenes and resolutions.

[0108] This invention employs two internationally recognized objective evaluation metrics: Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM). These metrics are widely used in image quality assessment and can objectively reflect the impact of image processing algorithms on image quality, measuring the algorithm's ability to enhance image details and preserve structural information. PSNR primarily measures pixel-domain information, while SSIM focuses on structural similarity, better reflecting the image's ability to maintain structure during detail enhancement. The image recognition results obtained using the OOA algorithm in this invention are optimal. Statistical data in the table show that the algorithm performed best in all SSIM tests, demonstrating its ability to maximize the preservation of image structural integrity while enhancing details. Furthermore, the algorithm also achieved leading rankings in most PSNR tests, further indicating its effective enhancement of image details within the pixel domain. This demonstrates that the proposed algorithm prioritizes the preservation of structural information during detail enhancement, resulting in superior visual effects, which is crucial for many practical applications. This invention is compared with existing algorithms, and the experimental results are shown in Table 1.

[0109] Table 1 Comparison of Image Enhancement and Recognition Results of Various Algorithm Models

[0110]

[0111] To more intuitively illustrate the technical effects of this invention, we compared the enhancement results of different algorithms through experiments, such as... Figure 3 and Figure 4 As shown. Figure 3 A magnified comparison diagram showing four times the detail of images of miners operating coal mining machines. Figure 3 The small image 'a' in the image is the original image of a miner operating a coal mining machine. Figure 3In the image, a1 is the original ground truth (GT); a2 is the image processed using the BFLS algorithm, with PSNR and SSIM values ​​of 27.88 dB and 0.8969, respectively; a3 is the image processed using the EGIF algorithm, with PSNR and SSIM values ​​of 28.28 dB and 0.9179, respectively; a4 is the image processed using the GGIF algorithm, with PSNR and SSIM values ​​of 27.61 dB and 0.877, respectively; and a5 is the image processed using the GIF algorithm, with PSNR and SSIM values ​​of [missing data]. The values ​​for SSIM in the numerical evaluation are 31.91 dB and 0.9882, respectively; the inset a6 shows the image processed using the RGIF algorithm, with values ​​of 27.61 dB and 0.877 in the PSNR and SSIM numerical evaluations, respectively; the inset a7 shows the image processed using the TH algorithm, with values ​​of 28.13 dB and 0.9396 in the PSNR and SSIM numerical evaluations, respectively; the inset a8 shows the image processed using the OOA optimization algorithm of this application, with values ​​of 35.51 dB and 0.9988 in the PSNR and SSIM numerical evaluations, respectively.

[0112] Figure 4 A four-fold magnified comparison diagram of images used by miners to assess ore conditions. Figure 4 In the image hierarchy, b1 is the original ground truth (GT); b2 is the image processed using the BFLS algorithm, with PSNR and SSIM values ​​of 29.21 dB and 0.9128, respectively; b3 is the image processed using the EGIF algorithm, with PSNR and SSIM values ​​of 29.58 dB and 0.9357, respectively; b4 is the image processed using the GGIF algorithm, with PSNR and SSIM values ​​of 30.16 dB and 0.9584, respectively; and b5 is the image processed using the GIF algorithm, with PSNR and SSIM values ​​of 29.21 dB and 0.9128, respectively. The values ​​for SSIM in the numerical evaluation are 32.21 dB and 0.9867, respectively; the image in b6 is the image processed by the RGIF algorithm, with values ​​of 30.16 dB and 0.9584 in the numerical evaluation of PSNR and SSIM, respectively; the image in b7 is the image processed by the TH algorithm, with values ​​of 28.14 dB and 0.9319 in the numerical evaluation of PSNR and SSIM, respectively; the image in b8 is the image processed by the OOA optimization algorithm in this application, with values ​​of 33.14 dB and 0.9997 in the numerical evaluation of PSNR and SSIM, respectively.

[0113] While the RGIF method produces a relatively smooth image enhancement, its ability to capture details is insufficient, leading to the loss of some detail information, especially in areas with complex textures where it fails to properly restore the original features. Algorithms such as TH, EGIF, GGIF, and BFLS, on the other hand, cause abnormal brightness due to over-enhancement, affecting the naturalness of the image. In contrast, the OOA algorithm proposed in this invention performs exceptionally well in feature extraction and enhancement at the detail level, producing more natural results and significantly outperforming other comparative algorithms in terms of visual effect.

[0114] In summary, the experimental results demonstrate that the proposed algorithm has certain applicability in the complex environment of underground mines. Mine environments typically present challenges such as low illumination, high noise, and dust particles, posing significant difficulties for image processing. By introducing a global optimization strategy, the algorithm can match similar image patches over a wider range, reducing local optima issues caused by insufficient local features, and improving the restoration of image details to some extent. Experimental results show that this method has a positive effect on improving the visualization quality and detail clarity of mine images, providing a valuable technical approach for image processing in underground mine environments.

[0115] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined by the scope of the claims.

Claims

1. A method for enhancing details in mine images based on the Osprey optimization algorithm, characterized in that, Includes the following steps, Step 1: Construct an image detail enhancement model based on residual learning, and apply the energy function... Y (z) Redesign; In step one, the image detail enhancement model based on residual learning is described as follows: , , Where z is a region block, and c(z) is the offset. D (z) is the image The region block centered at z. For the feasible region of z, I e This represents the final image with enhanced detail. The magnification factor for the detail layer is obtained by calculating all image region blocks. D (z) Initial residual characteristics Res; Step 2: Extract the initial residual by comparing the scaled image with the original image; Step 3: Improve the matching method using the Osprey optimization algorithm to find more suitable detailed layer features, making it easier for the system to converge to the global optimum. In step three, a feature refinement mechanism is designed based on the improved matching method using the Osprey optimization algorithm. To find more suitable detail layer features, further refine the initial Res residual features, and finally obtain the detail layer... The solution converges to obtain the global optimal solution; The improved matching method based on the Osprey optimization algorithm specifically includes: S3.1 Random initialization of the osprey population S3.2 The energy value of each pixel in I1 and the energy value of its corresponding Osprey agent in image I2 are calculated by weighted summation of brightness, gradient, and texture features. The absolute values ​​of these differences are then used to obtain the fitness values ​​of the N Osprey agents corresponding to each pixel. After comparison Find the current best fitness value in I2. ; S3.3 Find a reference location for each Osprey agent. This serves as the target for the next move, thus updating the position. S3.4 Reference Position Update the position to which the next decision will be made; S3.5 will update the Once the cropping reaches the boundary, a fitness-based selection mechanism is introduced to calculate the fitness value of the new position. Filter for the optimal position; S3.6 will be the final position after iteration. Mapping onto the initial residual image H0, the weighted average value of each point is calculated. Remove the values ​​from the expanded boundary portion. Finally, the updated residual we obtained is ; Step four: Manually adjust the magnification factor parameter to magnify the obtained detail layer, and then overlay it with the input image I0 to obtain the image I after detail enhancement. e .

2. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, The energy function Y (z), considering pixel loss and two regularization terms and , , , , The energy function Y (z) is a new block matching criterion, which, after optimization, yields c(z), i.e., c(z) = argmin. Y (z); and Each represents a certain region block D Hamiltonian and Laplacian operators for (z), Let z be the feasible region. and This represents the regularization constant that controls the prior components.

3. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, In step S3.2, the energy value of each pixel in I1 is subtracted from the energy value at the corresponding position in I2, and the absolute value is taken to obtain the fitness value. Compare the N Osprey agents corresponding to each pixel in I1. Find the best current fitness value in I2. And let its position be considered the "best" position, that is, the set of pixel positions with the smallest fitness value, denoted as , , Where N is the number of ospreys in the population, and t is the number of iterations. r t p N Osprey agents corresponding to each pixel The index of the minimum value individual. The coordinates of the indexed individual.

4. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, In step S3.3, for the first image in image I1... p The pixel at the th position t In each round of iteration, the assigned N search agents search for a set of agents in image I2 that have a better fitness than themselves. By combining the selection of neighborhood solutions and global optimal solutions, the particle is guided to move in a more favorable direction, with a reference position. ,in In the set A randomly selected index in the middle. rand is a random number that follows a uniform distribution [0,1].

5. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, In step 3.4, in the first t In the nth iteration, the image I1... p The first pixel sent out i An agent at the current location Based on this, the location is updated. ,in, For the first i An agent is currently at the following location. As a guiding point, This is the optimal position among the current N agents. For localized guiding weights, For global guiding weights, I The random factor is rand(1,dim), which is a random perturbation term, represented as a randomly generated matrix of size 1×dim.

6. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, In step S3.5, the new position The energy value and the corresponding pixel in image I1 p The fitness value for the new location is calculated by subtracting the energy values ​​and taking the absolute value. ,according to Select the current optimal value through comparison. ,Right now , ,if < Then the global optimum is updated to The global optimal position is also updated accordingly. ,otherwise Then assign for , for That is, at the start of the next iteration, the fitness value and position of the first Osprey agent will be forcibly changed to and Together with the other N-1 osprey agents, they form a new population of N.

7. The method for enhancing details in mine images based on the Osprey optimization algorithm according to claim 1, characterized in that, In step 3.6, the cumulative matrix is ​​updated by accumulating and superimposing. sumh 0 and counting matrix counth 0, calculate the weighted average of each point. .

Images