Exponential-trigonometric function inspired single image robot vision image detail enhancement algorithm

Abstract

This invention discloses a single-image robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions, comprising the following steps: Step 1, obtaining initial residual features from the original image; Step 2, designing a loss function to describe pixel value loss and evaluate whether the residual image patch matching reaches the optimal level; Step 3, initializing image candidate patches, setting search boundaries and the initial position of each image candidate patch, and comparing each image candidate patch with the designed loss function to obtain the initial optimal image patch and the initial optimal fitness; Step 4, using an optimization algorithm to iteratively search the original image using the initial residual features and perform residual updates to obtain the final optimal image patch; Step 5, superimposing the optimal image patch onto the original image to obtain the final detail-enhanced image. This invention can significantly improve the texture clarity and edge discernibility of robot-acquired images, directly improving the accuracy and stability of downstream visual tasks.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions. Background Technology

[0002] In today's rapidly developing robotics technology, robot vision systems, as a crucial component of robots' perception of their external environment, are core technologies for achieving autonomous localization and navigation, environmental perception, and human-robot interaction. With the widespread application of various robots in complex scenarios, high-quality visual data has become key to ensuring the accuracy of robot decision-making. Applying efficient image detail enhancement techniques to robot vision can significantly improve the perception accuracy of robots in complex environments, providing innovative solutions for various tasks such as Simultaneous Localization and Mapping (SLAM), object detection, and robotic arm grasping.

[0003] In actual robot operations, we often face complex and ever-changing lighting conditions and dynamic environmental challenges. For example, when inspection robots or home service robots move between indoor and outdoor environments, they often encounter complex situations such as drastic changes in lighting, low light, or motion blur, leading to loss of image details, low contrast, or severe noise interference. Traditional image enhancement methods (such as Laplacian sharpening and histogram equalization) often struggle to balance enhancement intensity with image naturalness when processing such complex visual data. They are prone to over-amplifying thermal noise from camera sensors or producing ringing effects and artifacts at object edges. These directly interfere with feature point extraction in visual odometry, causing robot positioning drift or mapping failure. Furthermore, the images processed by robots are highly diverse, with each type of detail feature and noise varying slightly. Traditional detail enhancement methods often cannot accommodate all aspects, resulting in image enhancement effects that fail to meet practical needs.

[0004] With advancements in deep learning technology, many deep learning methods based on Convolutional Neural Networks (CNNs) have demonstrated outstanding performance in image enhancement. However, their deployment on robotic end-users still faces significant challenges. Currently, commonly used robots typically rely on battery power and have extremely limited onboard computing resources, making it difficult to support real-time inference for highly complex deep learning models. Furthermore, the high computational cost significantly reduces the robot's battery life. In addition, deep learning methods depend on massive amounts of labeled data for training, but in real-world robotic applications, obtaining high-quality paired image data and accurately labeling it is extremely difficult and costly.

[0005] Current image processing technologies used in robot vision, especially in the field of image detail enhancement, while offering various solutions, still present some problems and challenges: (1) The balance between noise and detail preservation: In the process of image detail enhancement, how to accurately distinguish between details and noise is a core challenge. In robot vision systems, the cameras mounted on robots are often limited by cost or size, resulting in more noise in low-light or high-dynamic environments. Existing enhancement algorithms often face a dilemma: either to highlight the texture details of obstacles or landmarks, background noise is amplified simultaneously, interfering with the feature matching of the visual SLAM system; or to suppress noise, the image is over-smoothed, resulting in the loss of key edge information. In real-world working environments with rich textures but full of interference, how to ensure that the enhancement algorithm can both recover the key geometric details used for navigation and effectively avoid the interference of noise on the visual algorithm is an urgent problem to be solved.

[0006] (2) Dependence on computing resources and training data: Robot vision systems have extremely high requirements for real-time performance and often run on embedded platforms with limited computing power. Existing deep learning image enhancement solutions face two major bottlenecks: First, the limitations of computing power and power consumption. Complex neural network models are difficult to achieve low-latency real-time processing on mobile robots with limited computing resources, and will accelerate power consumption. Second, the data acquisition barrier. In the unstructured open world, it is extremely difficult to collect perfectly matched training data covering various extreme lighting and blurry scenes and perform pixel-level annotation.

[0007] The challenges in addressing these technical issues—balancing noise and detail preservation, and the dependence of computational resources on training data—pose significant obstacles in robot vision enhancement. Noise and detail are often indistinguishable, and the robot's motion causes image blurring and rapid changes in lighting, making image processing algorithm optimization difficult. Existing technologies often prioritize either noise suppression or detail preservation, making it difficult to achieve both simultaneously. While deep learning methods have shown promising results, their massive parameter count makes lightweight deployment on robot edge computing devices, which are extremely sensitive to real-time performance and power consumption. Summary of the Invention

[0008] The purpose of this invention is to provide an exponential-trigonometric function-inspired single-frame robot visual image detail enhancement algorithm, which can significantly improve the texture clarity and edge discernibility of robot-acquired images, directly improving the accuracy and stability of downstream visual tasks. At the same time, by reducing the dependence on high-performance computing hardware, it can significantly reduce the manufacturing cost and energy consumption of robots, extend operation time, and promote the development of robot technology towards a lighter, more intelligent, and more universal direction.

[0009] To achieve the above objectives, this invention provides a single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions, characterized by comprising the following steps: Step 1: Obtain the initial residual features from the original image; Step 2: Design a loss function to describe pixel value loss and evaluate whether the residual image patch matching has reached the optimal level; Step 3: Initialize image candidate blocks, set the search boundary and the initial position of each image candidate block, substitute each image candidate block into the designed loss function for comparison, and obtain the initial optimal image block and the initial optimal fitness. Step 4: Use an optimization algorithm to iteratively search the original image using the initial residual features and update the residuals to obtain the final optimal image patch; Step 5: Overlay the best image patch onto the original image to obtain the final detail-enhanced image.

[0010] As a further aspect of the present invention: the initial residual feature acquisition step in step one is as follows: First, input the original image, find a pixel position in the original image, and take this pixel position as the original image block. Then, calculate the pixel difference between all image blocks within a local range and the original image block, centered on this pixel position. Then, take the smallest pixel position as the offset to obtain the position offset closest to the original image block. Finally, repeat the above process for each pixel of the original image to obtain the initial residual features. The process is as follows: ; in, Indicates the location of the image patch. Indicates the offset. Representing an image China and Israel The image patch centered on Indicates a local area. This represents the initial residual characteristics.

[0011] As a further aspect of the present invention: the loss function in step two is specifically expressed as follows: ; in, This represents the loss function.

[0012] As a further aspect of the present invention: in step three, the image candidate blocks are the positions of randomly generated image blocks within the search range. The initialization of the image candidate blocks is initiated by randomly generating a set of potential image candidate blocks. Each image block position is represented by a two-dimensional vector, and the formula for the initial position of the image block is expressed as: ; in, Indicates the first The pixel in the first Position in the round of iteration, This represents a distribution in the interval [0,1]. random matrix, Indicates the lower limit of the search. Indicates the upper limit of the search; This stage generates N Each image candidate block is fed into the fitness function to calculate the initial fitness of each candidate block. The initial fitness of the smallest candidate block is selected as the initial optimal fitness, and this position is retained as the position of the initial optimal image block. The specific initial process is as follows: ; ; in, Indicates the optimal fitness; Indicates the position of the current optimal image patch; F ( Xi,j ) represents the fitness function.

[0013] As a further aspect of the present invention: the optimization algorithm in step four introduces a constraint search strategy to accelerate the search process, specifically as follows: ; ; in, Indicates the current iteration number. Indicates the total number of iterations. This represents the function in MATLAB that performs floor function rounding. and These represent adjustment coefficients, Indicates the number of iterations required to initiate the current and subsequent constraint exploration methods; When the number of iterations reaches The search upper limit U and search lower limit L will be adjusted accordingly to speed up the search. The specific adjustments to the search upper limit U and search lower limit L are as follows: ; ; in, and This represents two coefficients randomly selected between 0 and 1. Indicates the position of the current optimal image patch. This indicates the position of the current suboptimal image patch.

[0014] As a further aspect of the present invention: ; in, This represents the critical number of iterations required for the transition between two stages; In the first exploration phase, potential locations are searched within the original image area. Image candidate blocks move towards finding the optimal pixel location. The update of the image candidate blocks is specifically as follows: ; in, These represent the positions of the y-th pixel in the current iteration and subsequent iterations, respectively. and Represents a random number within the interval [0,1]. The weights for controlling candidate solutions in the first exploration phase are as follows: ; Introducing parameters and The search space is created where these two values ​​decrease with increasing iterations, and they are symmetric functions, specifically: ; In the second exploration phase, the movement of image candidate blocks is determined by the current position of the image candidate block, specifically: ; in, Represents a random number in the range [0,1]. The weights for controlling candidate solutions in the second exploration phase are specifically expressed as follows: ; The transition between the exploration and development phases is as follows: ; in, and Calculated using equation (13), when A value greater than 1 indicates the exploration phase, while a value less than 1 indicates the development phase. In the first development phase, the focus is on exploring the surrounding area of ​​the current optimal image patch location, specifically: ; in, and A random number in the range [0,1]. The weights for controlling candidate solutions in the first development phase are as follows: ; In the second development phase, the candidate solutions controlled in the first development phase are further explored around the already determined optimal solution. This enhanced search strategy specifically involves: ; ; in, is a coefficient used to maintain the diversity of candidate solutions.

[0015] As a further aspect of the present invention: in step five, the optimal residual features are superimposed onto the input image, specifically as follows: ; in, For the enhanced image, As an enhancing factor, An optimization algorithm for updating residual features, This represents the updated residual features.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention analyzes the self-similarity of residual details in images, enabling adaptive adjustment of enhancement strategies based on image characteristics. This ensures natural enhancement of details, avoids excessive noise amplification, and meets the needs of robots operating in multiple scenarios. It innovatively introduces an exponential-trigonometric function optimization algorithm, effectively improving the detail representation of images acquired by robots. This provides a new technical means for the field of image processing and has broad application prospects in the construction of intelligent perception for robots. Attached Figure Description

[0017] Figure 1 This is a flowchart of the algorithm of the present invention; Figure 2 This is a flowchart of the algorithm of the present invention; Figure 3 Subjective comparison of the algorithms Figure 1 ; Figure 4 Subjective comparison of the algorithms Figure 2 ; Figure 5 Subjective comparison of the algorithms Figure 3 ; Figure 6 To quantify the comparison results; Figure 7 This is the algorithm's pseudocode. Detailed Implementation

[0018] The present invention will be further illustrated by the following examples.

[0019] This invention utilizes statistical information to extract local residual features from images, then updates and optimizes these features to obtain detailed features. Detail enhancement is achieved by finding the image patch with the most similar pixel values ​​to each image block in the image; the smaller the pixel difference, the richer the image detail. In robot vision processing, many details in the environment (such as carpet texture, wall edges, and object surface materials) exhibit significant scale dependence and self-similarity. Accurate matching of image patches can effectively reveal these complex features. For example, in visual SLAM, stable extraction of feature points depends on clear texture structures; accurate matching of residual blocks can better capture this information, improving the accuracy of detail enhancement. By finding the minimum pixel difference, not only can important geometric details in the image be enhanced, but the naturalness and realism of the details can also be ensured, avoiding artificial artifacts. For instance, in robot obstacle detection, enhancing image details can better identify small obstacles or terrain differences, providing effective support for path planning.

[0020] This invention leverages the inherent nonlocal self-similarity of images to extract high-frequency residuals through multi-scale layering. To address the inefficiencies and susceptibility to local optima inherent in traditional search methods, this invention innovatively introduces an exponential-trigonometric function optimization algorithm to search for the best matching block. This invention transforms the image detail restoration problem into a location optimization problem. The exponential-trigonometric function optimization algorithm first constructs a high-frequency residual layer and a low-frequency approximation layer of the image. Then, for each image block to be restored, the exponential-trigonometric function optimization algorithm iterates rapidly within the search space to find the location of the most similar low-frequency block. Finally, the high-frequency information corresponding to this location is mapped back to the original image, thereby achieving detail enhancement.

[0021] like Figure 1 As shown, the single-image robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions of this invention includes the following steps: Step 1: Obtain the initial residual features from the original image; Specifically, the initial residual feature acquisition steps are as follows: First, input the original image and find a pixel location in the original image. This pixel location is taken as the original image block. Then, using this pixel location as the center, calculate the pixel difference between all image blocks within a local range and the original image block. Then, take the pixel location with the smallest difference as the offset to obtain the position offset closest to the original image block. Finally, repeat the above process for each pixel of the original image to obtain the initial residual features. The process is as follows: ; in, Indicates the location of the image patch. Indicates the offset. Representing an image China and Israel The image patch centered on Indicates a local area. This represents the initial residual characteristics.

[0022] Step 2: For image detail enhancement, Steps 3 and 4 are used for search optimization. For optimal image matching, a loss function is designed to describe pixel value loss and evaluate whether the residual image patch matching has reached the optimal level. Specifically, the loss function is expressed as follows: ; in, This represents the loss function.

[0023] Step 3: Initialize image candidate blocks, set the search boundary and the initial position of each image candidate block, substitute each image candidate block into the designed loss function for comparison, and obtain the initial optimal image block and the initial optimal fitness. Specifically, image candidate blocks are the locations of randomly generated image blocks within the search range. Initialization of image candidate blocks is initiated by randomly generating a set of potential image candidate blocks. Each image block location is represented by a two-dimensional vector, and the formula for initializing the image block location is expressed as: ; in, Indicates the first The pixel in the first Position in the round of iteration, This represents a distribution in the interval [0,1]. random matrix, Indicates the lower limit of the search. Indicates the upper limit of the search.

[0024] The initialization phase also includes generating initial optimal image patches. This invention uses pixel loss as the loss function; smaller pixel differences indicate better system performance. The residuals are updated using the loss function from step two. First, the initial fitness is set to infinity, and then the images generated in this phase... N Each image candidate block is fed into the fitness function to calculate the initial fitness of each candidate block. The initial fitness of the smallest candidate block is selected as the initial optimal fitness, and this position is retained as the position of the initial optimal image block. The specific initial process is as follows: ; ; in, Indicates the optimal fitness; Indicates the position of the current optimal image patch; F ( Xi,j ) represents the fitness function.

[0025] Step 4: Use an optimization algorithm to iteratively search the original image using the initial residual features and update the residuals to obtain the final optimal image patch; The optimization algorithm includes the following steps: The optimization algorithm of this invention introduces a constraint search strategy to accelerate the search process, specifically as follows: ; ; in, Indicates the current iteration number. Indicates the total number of iterations. This represents the function in MATLAB that performs floor function rounding. and These represent adjustment coefficients, Indicates the number of iterations required to initiate the current and subsequent constraint exploration methods; When the number of iterations reaches The search upper limit U and search lower limit L will be adjusted accordingly to speed up the search. The specific adjustments to the search upper limit U and search lower limit L are as follows: ; ; in, and This represents two coefficients randomly selected between 0 and 1. Indicates the position of the current optimal image patch. This indicates the position of the current suboptimal image patch.

[0026] Furthermore, the optimization process of the optimization algorithm includes an exploration phase and a development phase. Each of the exploration and development phases comprises two stages: a first exploration stage, a second exploration stage, a first development stage, and a second development stage. The two stages of the exploration phase can be converted into: ; in, This represents the critical number of iterations required for the transition between two stages; In the first exploration phase, potential locations are searched within the original image area. Image candidate blocks move towards finding the optimal pixel location. The update of the image candidate blocks is specifically as follows: ; in, These represent the positions of the y-th pixel in the current iteration and subsequent iterations, respectively. and Represents a random number within the interval [0,1]. The weights for controlling candidate solutions in the first exploration phase are as follows: ; Introducing parameters and The search space is created where these two values ​​decrease with increasing iterations, and they are symmetric functions, specifically: ; In the second exploration phase, the movement of image candidate blocks is determined by the current position of the image candidate block, specifically: ; in, Represents a random number in the range [0,1]. The weights for controlling candidate solutions in the second exploration phase are specifically expressed as follows: ; The transition between the exploration and development phases is as follows: ; in, and Calculated using equation (13), when A value greater than 1 indicates the exploration phase, while a value less than 1 indicates the development phase. In the first development phase, the focus is on exploring the surrounding area of ​​the current optimal image patch location, specifically: ; in, and A random number in the range [0,1]. The weights for controlling candidate solutions in the first development phase are as follows: ; In the second development phase, the candidate solutions controlled in the first development phase are further explored around the already determined optimal solution. This enhanced search strategy specifically involves: ; ; in, is a coefficient used to maintain the diversity of candidate solutions.

[0027] Step 5: The optimal residual features obtained from the search are superimposed onto the input image to obtain the final detail-enhanced image; Specifically, the optimal residual features are superimposed onto the input image, which is represented as follows: ; in, For the enhanced image, As an enhancing factor, An optimization algorithm for updating residual features, This represents the updated residual features.

[0028] The algorithm framework of this invention is as follows: Figure 2 As shown, the algorithm consists of three modules: residual initialization, residual optimization, and image enhancement.

[0029] 1) The residual initialization module obtains the initial residual characteristics through bilinear interpolation.

[0030] 2) Starting from pixel loss, the residual optimization module continuously updates the pixel position through two stages: exploration and development of exponential-trigonometric function optimization, until the maximum number of iterations is reached. It dynamically searches for the optimal matching block, continuously approaches the optimal point of the residual space, and obtains the final optimized residual features.

[0031] 3) The image enhancement module first treats the final residual features obtained in the above steps as the detail features of the image, then multiplies them by an enhancement factor, and then linearly superimposes them onto the original input image to obtain the final detail-enhanced image.

[0032] The overall process of the algorithm is as follows: Algorithm pseudocode is as follows Figure 7 As shown.

[0033] To visually demonstrate the effectiveness of the proposed algorithm, this invention provides a comparison with several other image detail enhancement algorithms. The algorithms compared include GIF, WGIF, BFLS, TH, ILS, and IPRH, which represent common methods in the current field of image enhancement. To verify the superiority of the proposed algorithm, this invention provides subjective visual effect comparison images and objective numerical evaluation metrics on classic datasets compared with other algorithms. To objectively evaluate image quality, we selected the internationally accepted Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) as evaluation criteria, and Equations (22) and (23) are their calculation methods. PSNR is used to measure the noise and distortion of an image, while SSIM evaluates the structural fidelity of the image. The combination of these two can more comprehensively reflect image quality.

[0034] ; ; in, It is the maximum pixel value in the processed image. It is the mean square error. and The average brightness values ​​of the original image and the image to be calculated. and These are the variances of the original image and the image to be calculated, respectively. It is the covariance between the original image and the image to be calculated. and It is a near-zero constant term that has no practical meaning and serves a regulating function.

[0035] Figure 3 , Figure 4 and Figure 5 Three example images of robot vision acquisition are shown, among which... Figure 5 This is an image of a robot working underground in a mine. In the comparison image, the algorithm proposed in this invention is highlighted in bold. Visually, overall, the algorithm proposed in this invention shows relatively ideal results in enhancing image details and avoiding excessive noise amplification. Other enhancement algorithms, to varying degrees, suffer from problems such as over-enhancement leading to image distortion and overexposure. Figure 4 Other algorithms used in this study clearly show that the image exhibits mosaic-like artifacts and severe color distortion in certain areas. The brightness on the horse's back is excessively high, further exacerbating the mosaic effect. Furthermore, large areas of other colors appear on the grass, severely limiting the reliable application of visual algorithms in robot environmental perception. In contrast, this invention not only provides excellent detail enhancement but also minimizes structural alterations to the original image, achieving maximum values ​​in both PSNR and SSIM evaluations.

[0036] To further illustrate the effectiveness of the algorithm, this invention also provides quantitative comparison results with other algorithms on different datasets, including BSDS200, RealSRSet, and T91. The algorithms compared include GIF, GGIF, WGIF, ZF, BFLS, TH, ILS, IPRH, CSGIS, PTF, QWLS, and ALSP. Figure 6 As shown. Figure 6 The values ​​in bold are the maximum values, and the values ​​in underlined text are the suboptimal values. It can be seen that the proposed algorithm has the best overall numerical performance on all three datasets.

[0037] In summary, the proposed algorithm demonstrates powerful image enhancement capabilities, resulting in clearer details and more natural visual effects in visual images acquired by the robot. For high-contrast, color-rich images, the algorithm effectively preserves color saturation and detail levels, avoiding distortion caused by over-enhancement. Comparative results show that the algorithm not only improves image quality but also effectively preserves key details, demonstrating greater potential. Compared to traditional methods, this algorithm outperforms traditional methods in preserving image details and structure, avoiding color distortion, excessive brightness amplification, and severe artifacts. Through a residual optimization mechanism, the algorithm enhances visual effects while preserving the original image structure.

[0038] The exponential-trigonometric function-inspired detail enhancement method of this invention not only has significant academic value but also holds great potential in practical robot operations. It can further improve the quality and processing efficiency of image enhancement, providing more accurate data support for robot perception systems and autonomous navigation. In the field of robot vision, the algorithm effectively solves the adaptability problem of traditional image enhancement techniques in complex dynamic environments, significantly improving the image analysis capabilities of robots in areas such as SLAM (Simultaneous Localization and Intervention), target recognition, obstacle avoidance, and precision operation, powerfully promoting the development of robots towards higher intelligence, higher efficiency, and safer operation.

Claims

1. A single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions, characterized in that, Includes the following steps: Step 1: Obtain the initial residual features from the original image; Step 2: Design a loss function to describe pixel value loss and evaluate whether the residual image patch matching has reached the optimal level; Step 3: Initialize image candidate blocks, set the search boundary and the initial position of each image candidate block, substitute each image candidate block into the designed loss function for comparison, and obtain the initial optimal image block and the initial optimal fitness. Step 4: Use an optimization algorithm to iteratively search the original image using the initial residual features and update the residuals to obtain the final optimal image patch; Step 5: Overlay the best image patch onto the original image to obtain the final detail-enhanced image.

2. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 1, characterized in that, Step 1, initial residual feature acquisition steps: First, input the original image, find a pixel position in the original image, and take this pixel position as the original image block. Then, calculate the pixel difference between all image blocks within a local range and the original image block, centered on this pixel position. Then, take the pixel position with the smallest difference as the offset to obtain the position offset closest to the original image block. Finally, repeat the above process for each pixel of the original image to obtain the initial residual features. The process is as follows: ； in, Indicates the location of the image patch. Indicates the offset. Representing an image China and Israel The image patch centered on Indicates a local area. This represents the initial residual characteristics.

3. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 2, characterized in that, The loss function in step two is specifically expressed as follows: ； in, This represents the loss function.

4. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 3, characterized in that, In step three, the image candidate blocks are the positions of randomly generated image blocks within the search range. Initialization of the image candidate blocks begins by randomly generating a set of potential image candidate blocks. Each image block position is represented by a two-dimensional vector, and the formula for initializing the image block position is as follows: ； in, Indicates the first The pixel in the first Position in the round of iteration, This represents a distribution in the interval [0,1]. random matrix, Indicates the lower limit of the search. Indicates the upper limit of the search; This stage generates N Each image candidate block is fed into the fitness function to calculate the initial fitness of each candidate block. The initial fitness of the smallest candidate block is selected as the initial optimal fitness, and this position is retained as the position of the initial optimal image block. The specific initial process is as follows: ； ； in, Indicates the optimal fitness; Indicates the position of the current optimal image patch; F ( Xi,j ) represents the fitness function.

5. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 4, characterized in that, In step four, the optimization algorithm introduces a constraint search strategy to accelerate the search process, specifically as follows: ； ； in, Indicates the current iteration number. Indicates the total number of iterations. This represents the function in MATLAB that performs floor function rounding. and These represent adjustment coefficients, Indicates the number of iterations required to initiate the current and subsequent constraint exploration methods; When the number of iterations reaches The search upper limit U and search lower limit L will be adjusted accordingly to speed up the search. The specific adjustments to the search upper limit U and search lower limit L are as follows: ； ； in, and This represents two coefficients randomly selected between 0 and 1. Indicates the position of the current optimal image patch. This indicates the position of the current suboptimal image patch.

6. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 5, characterized in that, The optimization process includes an exploration phase and a development phase. Each of the exploration and development phases contains two stages: a first exploration phase and a second exploration phase, as well as a first development phase and a second development phase. The two stages of the exploration phase are converted into: ； in, This represents the critical number of iterations required for the transition between two stages; In the first exploration phase, potential locations are searched within the original image area. Image candidate blocks move towards finding the optimal pixel location. The update of the image candidate blocks is specifically as follows: ； in, These represent the positions of the y-th pixel in the current iteration and subsequent iterations, respectively. and Represents a random number within the interval [0,1]. The weights for controlling candidate solutions in the first exploration phase are as follows: ； Introducing parameters and The search space is created where these two values ​​decrease with increasing iterations, and they are symmetric functions, specifically: ； In the second exploration phase, the movement of image candidate blocks is determined by the current position of the image candidate block, specifically: ； in, Represents a random number in the range [0,1]. The weights for controlling candidate solutions in the second exploration phase are specifically expressed as follows: ； The transition between the exploration and development phases is as follows: ； in, and Calculated using equation (13), when A value greater than 1 indicates the exploration phase, while a value less than 1 indicates the development phase. In the first development phase, the focus is on exploring the surrounding area of ​​the current optimal image patch location, specifically: ； in, and A random number in the range [0,1]. The weights for controlling candidate solutions in the first development phase are as follows: ； In the second development phase, the candidate solutions controlled in the first development phase are further explored around the already determined optimal solution. This enhanced search strategy specifically involves: ； ； in, is a coefficient used to maintain the diversity of candidate solutions.

7. The single-frame robot visual image detail enhancement algorithm inspired by exponential-trigonometric functions according to claim 6, characterized in that, In step five, the optimal residual features are superimposed onto the input image, specifically as follows: ； in, For the enhanced image, As an enhancing factor, An optimization algorithm for updating residual features, This represents the updated residual features.

Images