A homomorphic filtering-CLAHE remote sensing image enhancement method integrated with strategy multi-objective particle swarm optimization

Abstract

This invention relates to the field of remote sensing image processing technology, specifically to a homomorphic filtering-CLAHE remote sensing image enhancement method based on integrated strategy multi-objective particle swarm optimization. The method includes: acquiring remote sensing image data and preprocessing it; converting the preprocessed remote sensing image to the HSV color space and extracting the V component; constructing a frequency-spatial hybrid enhancement framework combining homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component; constructing an integrated strategy multi-objective particle swarm optimization algorithm; and constructing a four-objective fitness function including structural similarity, average gradient, information entropy, and gray-level variance, using this function to guide particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters. This invention can effectively enhance the clarity and structural integrity of terrain texture in remote sensing images of complex mountainous areas.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and more specifically, to a homomorphic filtering-CLAHE remote sensing image enhancement method based on integrated strategy multi-objective particle swarm optimization. Background Technology

[0002] Existing remote sensing image enhancement methods each have their limitations: Histogram equalization (HE), while improving global contrast, is prone to local over-enhancement and artifacts; Adaptive Histogram Equalization (AHE) and Improved Contrast-Limited Adaptive Histogram Equalization (CLAHE) improve details through local processing, but still suffer from noise amplification or artifacts in flat areas; Wavelet Transform (WT) struggles to capture anisotropic features when separating high- and low-frequency information, and its effectiveness in handling complex features is limited; Retinex algorithms and multi-scale variants are prone to producing halos at edges and amplifying noise; Gamma correction (such as Local Adaptive Gamma Correction (LAGC)) tends to lose details or amplify noise in bright or low-light areas; Homomorphic Filtering (HF) can correct uneven illumination, but its parameters rely on manual adjustment, which can easily lead to blurred details. Traditional parametric methods (such as HF and CLAHE) rely on manual experience to select parameters, making it difficult to guarantee optimal results in complex scenes and lacking adaptability. While swarm intelligence optimization algorithms (such as particle swarm optimization, PSO) can achieve parameter optimization, traditional PSO is a single-objective optimizer. It needs to transform multiple objectives into a single objective through weighting. The weight allocation depends on prior knowledge and cannot guarantee the diversity of solution sets. Although multi-objective particle swarm optimization (MOPSO) introduces the concept of Pareto optimality, it still has room for improvement in balancing convergence and diversity. It is difficult to simultaneously meet the multi-objective optimization needs of image contrast, brightness balance and detail preservation. Summary of the Invention

[0003] In view of this, in order to address the shortcomings of the prior art, the present invention proposes a homomorphic filtering-CLAHE remote sensing image enhancement method with integrated strategy multi-objective particle swarm optimization, aiming to solve at least one of the problems mentioned in the background art.

[0004] This invention provides an integrated strategy multi-objective particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method, comprising the following steps: Acquire remote sensing image data and preprocess it, including radiometric calibration, atmospheric correction and spatial cropping; The preprocessed remote sensing image is converted to the HSV color space, and the V channel component is extracted. A frequency-spatial hybrid enhancement framework combining homomorphic filtering and contrast-limited adaptive histogram equalization is constructed, and this framework is used to enhance the V channel components. An integrated strategy multi-objective particle swarm optimization algorithm is constructed to optimize the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework. A four-objective fitness function, including structural similarity, average gradient, information entropy, and gray variance, is constructed. This function guides particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters.

[0005] In some embodiments, when constructing the frequency-spatial hybrid enhancement framework of homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component, the following steps are included: image Decomposed into low-frequency irradiation components and high-frequency reflection components : Perform a logarithmic transformation on the image to separate it into low-frequency illumination. With high frequency reflection : Performing a Fourier transform yields the frequency domain expression: in, and These represent the frequency domain components of the illumination and reflection components, respectively; In the frequency domain space Perform Gaussian filtering : Reconstructed and enhanced spatial image: The V-channel components are filtered using an optimized Gaussian homomorphic filter, including: The transfer function of the optimized Gaussian homomorphic filter is: ; in, , These represent high-frequency gain and low-frequency gain, respectively. and Separately control high-frequency enhancement and low-frequency attenuation; This represents the distance from a point in the frequency domain to the center. Image size, The cutoff frequency, This is the steepness parameter of the transition zone.

[0006] In some embodiments, when constructing the frequency-spatial hybrid enhancement framework of homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component, the method further includes: A contrast-limited adaptive histogram equalization algorithm is applied to the V channel component after homomorphic filtering. This algorithm includes dividing the image into sub-blocks, truncating and redistributing the histogram of each sub-block, generating a normalized cumulative distribution function and mapping grayscale values, and interpolating the mapping functions of adjacent sub-blocks. Image Divided into Each sub-block histogram Defined as: in, For gray levels, It is the Dirac function; Introduce a contrast limiting threshold Truncate the histogram for each sub-block: Where M is the number of pixels in the sub-block; After truncation, the histogram maintains the total number of pixels through linear redistribution: Generate normalized CDF And map grayscale values: Interpolate the mapping function of adjacent sub-blocks to output pixel values. for: Among them, weight Determined by pixel position.

[0007] In some embodiments, when constructing the dynamic cooperative multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm includes: The inertia weight and learning factor are dynamically adjusted using the Sigmoid function, where the dynamic change formula for the inertia weight is: In the formula: , They are 0.9 and 0.4 respectively; For the Sigmoid function; Indicates the current iteration number; , Indicates the maximum number of iterations; The cognitive coefficient C1, social coefficient C2, and neighborhood coefficient C3 are dynamically adjusted using the Sigmoid function, as shown in the following formula: In the formula: Set to 2; ; , Set them to 2.5 and 0.5 respectively; ; Set it to 1.5.

[0008] In some embodiments, when constructing the integrated strategy multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm further includes: constructing a dynamic random neighborhood topology, wherein in each iteration, each particle... neighborhood Generated through random sampling, including: In the formula: Describes the particle in the t-th iteration. The neighborhood set; N Indicates the size of the particle swarm; Indicates the size of the neighborhood; Represents a uniformly random permutation function; in the neighborhood Internally, a dominant relationship is adopted. Choose the solution with the strongest dominance as... If there is no dominant solution, then a non-dominated solution is randomly selected.

[0009] In some embodiments, when constructing the integrated strategy multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm further includes: designing adaptive Gaussian mutation, calculating the normalized difference between the fitness of the current particle and the global optimal solution, dynamically adjusting the mutation probability and mutation amplitude based on the difference, and performing Gaussian perturbation on the particle position, including: the formula for calculating the normalized difference between the fitness of the current particle and the global optimal solution is: in: It is the total number of dimensions of the objective function; It is the dimension of the objective function; It is the fitness vector of the global optimal solution; It is the fitness vector of the current particle; the denominator is... It is the globally optimal k-dimensional absolute value; It is a local minimum. This ensures that attention is paid to dimensions where particle performance is insufficient; The formulas for calculating the probability of mutation and the magnitude of mutation are: in: , representing the basic mutation probability; , represents the adjustment coefficient; , indicating the basic range; , represents the magnification factor; Generate random numbers based on the above parameters. ,like Then a Gaussian mutation is performed, the mutation operation being defined as altering the particle position. Disturbance: in The standard normal random perturbation is standard Gaussian noise that is independent in each dimension.

[0010] In some embodiments, the construction of a four-objective fitness function including structural similarity, average gradient, information entropy, and gray-level variance, and the use of this function to guide particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters, includes: The fitness evaluation function formula is: in, This represents an optimization problem with four objectives. Indicators representing structural similarity; Indicates the average gradient; Represents information entropy; Indicates variance; This represents a parameter vector.

[0011] In some embodiments, the construction of a four-objective fitness function including structural similarity, average gradient, information entropy, and gray-level variance, and the use of this function to guide particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters, further includes: The formula for calculating structural similarity is: in, , Images , The average value; , Images , The variance; For image , covariance; , It is a constant; The formula for calculating the average gradient is: in, , The image at the pixel points Horizontal and vertical gradients at the location, and These represent the number of rows and columns of the image, respectively. The formula for calculating information entropy is: in, grayscale The probability of occurrence; The gray level of the image; The formula for calculating the variance of gray levels is: in, For the image at the pixel point The grayscale value at that location; This represents the average grayscale value of the image.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The remote sensing image enhancement under the frequency domain-spatial domain hybrid framework enhances both global brightness and local texture details. Compared with conventional HF, HE, WT methods, it is superior in enhancing the contrast, global brightness and terrain texture details of remote sensing images in complex mountainous areas.

[0013] (2) The ISMOPSO method, which integrates dynamic parameter adjustment, neighborhood guidance and adaptive mutation, is introduced. The multi-objective fitness evaluation model is used to determine the near-optimal parameter factors, effectively avoiding over-enhancement and halo artifacts. It can better reflect the complex terrain features of remote sensing images, especially in terms of structural integrity and detail representation in complex terrain areas.

[0014] (3) Compared with enhancement methods such as MOPSO-HC, HE, WT, SSR, LAGC, and HF, the standard deviation and average information entropy of the processed image were increased by 97.81% and 19.02% respectively, and the visual information fidelity, average gray level, spatial frequency and average gradient were increased by 4.25 times, 2.92 times, 2.44 times and 2.55 times respectively.

[0015] The above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0016] Other features and aspects of this disclosure will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 Flowchart of the integrated strategy multi-objective particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method provided in this embodiment of the invention; Figure 2 Flowchart of the integrated strategy multi-objective particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method provided in this embodiment of the invention; Figure 3 A three-dimensional structure diagram of the transfer function of the Gaussian homomorphic filter in the CLAHE remote sensing image enhancement method with integrated strategy multi-objective particle swarm optimization provided in this embodiment of the invention. Figure 4 A diagram illustrating the dynamic parameter adjustment strategy of the integrated strategy multi-target particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method provided in this embodiment of the invention; Figure 5 An example diagram illustrating the dynamic random neighborhood topology update process of the homomorphic filtering-CLAHE remote sensing image enhancement method with integrated strategy multi-objective particle swarm optimization provided in this embodiment of the invention. Figure 6 A comparison of the enhancement effects of different algorithms on remote sensing images provided by the integrated strategy multi-objective particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method in the embodiments of the present invention; Figure 7 Image structure autocorrelation results of the homomorphic filtering-CLAHE remote sensing image enhancement method with integrated strategy multi-objective particle swarm optimization provided in this embodiment of the invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] See Figure 1-7 As shown, an integrated strategy multi-objective particle swarm optimization homomorphic filtering-CLAHE remote sensing image enhancement method according to an embodiment of this application includes the following steps: Acquire remote sensing image data and preprocess it, including radiometric calibration, atmospheric correction and spatial cropping; The preprocessed remote sensing image is converted to the HSV color space, and the V channel component is extracted. A frequency-spatial hybrid enhancement framework combining homomorphic filtering and contrast-limited adaptive histogram equalization is constructed, and this framework is used to enhance the V channel components. An integrated strategy multi-objective particle swarm optimization algorithm is constructed to optimize the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework. A four-objective fitness function, including structural similarity, average gradient, information entropy, and gray variance, is constructed. This function guides particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters.

[0021] It should be understood that Shangri-La City (99°20′–100°19′E, 26°52′–28°52′N) is located in the core area of ​​the "Three Parallel Rivers" World Natural Heritage Site in the heart of the Hengduan Mountains. It is characterized by high altitude and low latitude, with a general topographic trend of high in the northwest and low in the southeast. The average altitude within the area is 3459m, with a maximum elevation difference of 4042m, making it a typical high-altitude, complex mountainous region. This extreme topography results in significant topographic shading effects in the images, and information about the mountain gullies is severely lacking.

[0022] The remote sensing imagery used in this study was obtained from the Landsat 8OLI / TIRSLevel-1 dataset, sourced from the official website of the U.S. Geological Survey (https: / / earthexplorer.usgs.gov / ), with a spatial resolution of 30 meters (OLI multispectral bands). Preprocessing was performed in ENVI 5.3, including radiometric calibration and atmospheric correction (FLAASH atmospheric correction). Natural true-color remote sensing images composed of Red, Green, and Blue bands were then spatially cropped to obtain a set of six 600-pixel × 600-pixel images, which served as the experimental dataset.

[0023] Table 1 Landsat 8 data used in the study

[0024] In some specific embodiments, when constructing the frequency-spatial hybrid enhancement framework of homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component, the following steps are included: image Decomposed into low-frequency irradiation components and high-frequency reflection components : Perform a logarithmic transformation on the image to separate it into low-frequency illumination. With high frequency reflection : Performing a Fourier transform yields the frequency domain expression: in, and These represent the frequency domain components of the illumination and reflection components, respectively; In the frequency domain space Perform Gaussian filtering : Reconstructed and enhanced spatial image: The V-channel components are filtered using an optimized Gaussian homomorphic filter, including: The transfer function of the optimized Gaussian homomorphic filter is: ; in, , These represent high-frequency gain and low-frequency gain, respectively. and Separately control high-frequency enhancement and low-frequency attenuation; This represents the distance from a point in the frequency domain to the center. Image size, The cutoff frequency, This is the steepness parameter of the transition zone.

[0025] It should be understood that homomorphic filtering is based on the illumination-reflection model, and the image... It can be decomposed into low-frequency irradiation components. and high-frequency reflection components .Right now: To separate and independently control the two types of components, a logarithmic transformation is first performed on the image to separate them into low-frequency illumination components. With high frequency reflection : Then, a Fourier transform is performed to obtain the frequency domain expression: in, and These represent the frequency domain components of the illumination and reflection components, respectively. In the frequency domain space... Perform Gaussian filtering To suppress low frequencies and enhance high frequencies: The filtered signal is then subjected to inverse Fourier transform and exponential restoration to reconstruct the enhanced spatial domain image. This study uses a Gaussian function as the transfer function for homomorphic filtering. The transfer function of a traditional Gaussian homomorphic filter is: The optimized Gaussian homomorphic filter transfer function is: in, and Separately control high-frequency enhancement and low-frequency attenuation; Represents the distance from the frequency domain point to the center ( (Image size); The cutoff frequency; This is the steepness parameter of the transition zone.

[0026] In some specific embodiments, when constructing the frequency-spatial hybrid enhancement framework of homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component, the method further includes: A contrast-limited adaptive histogram equalization algorithm is applied to the V channel component after homomorphic filtering. This algorithm includes dividing the image into sub-blocks, truncating and redistributing the histogram of each sub-block, generating a normalized cumulative distribution function and mapping grayscale values, and interpolating the mapping functions of adjacent sub-blocks. Image Divided into Each sub-block histogram Defined as: in, For gray levels, It is the Dirac function; Introduce a contrast limiting threshold Truncate the histogram for each sub-block: Where M is the number of pixels in the sub-block; After truncation, the histogram maintains the total number of pixels through linear redistribution: Generate normalized CDF And map grayscale values: Interpolate the mapping function of adjacent sub-blocks to output pixel values. for: Among them, weight Determined by pixel position.

[0027] It should be understood that this restricts contrast-adaptive histogram equalization; Contrast-Limited Adaptive Histogram Equalization (CLAHE) is a classic local contrast enhancement method, often used to improve image detail and effectively suppress noise amplification caused by traditional histogram equalization. Unlike global equalization methods, CLAHE divides the image into multiple small blocks and performs histogram equalization on each block, making the enhancement more locally adaptive.

[0028] Image Divided into Each sub-block histogram Defined as: in, For gray levels, This is the Dirac function.

[0029] To prevent noise amplification, a contrast limiting threshold is introduced. Truncate the histogram for each sub-block: M This represents the number of pixels in the sub-block. After truncation, the histogram maintains the total number of pixels through linear redistribution. Generate normalized CDF And map grayscale values: To avoid block artifacts, the mapping function of adjacent sub-blocks is interpolated. Taking a four-neighbor sub-block as an example, the output pixel value... for: Among them, weight Determined by pixel position.

[0030] In remote sensing image enhancement, the selection of image quality assessment indicators directly affects the search direction, convergence, and enhancement effect of the optimization algorithm. A single quality indicator is difficult to comprehensively evaluate the enhancement effect. Therefore, this study proposes a multi-objective fitness evaluation model that integrates structural similarity index (SSIM), average gradient, information entropy, and variance. This model quantifies image quality from different perspectives and guides the particles to search in the direction of simultaneously optimizing multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters. The fitness evaluation function is shown in Equation (13): in, This represents an optimization problem with four objectives. This represents a parameter vector.

[0031] SSIM (Structural Similarity Index) measures the similarity between two images in terms of structure, brightness, and contrast. The SSIM value ranges from [0,1], with higher values ​​indicating greater similarity. The calculation formula is as follows: in, , Images , The average value; , Images , The variance; For image , covariance; , It is a constant.

[0032] This represents the average gradient, reflecting the sharpness and richness of edge information in an image. A larger average gradient indicates sharper edges and richer details. The formula is as follows: in, , The image at the pixel points The horizontal and vertical gradients at a given point are calculated using the Sobel operator. and These represent the number of rows and columns of the image, respectively.

[0033] Information entropy measures the richness of information in an image. Higher information entropy indicates a more uniform grayscale distribution and a greater amount of information. Its calculation formula is: in, grayscale The probability of occurrence; represents the number of gray levels in the image.

[0034] Variance represents the degree of dispersion of image grayscale values. The larger the variance, the higher the image contrast.

[0035] in, For the image at the pixel point The grayscale value at that location; This represents the average grayscale value of the image.

[0036] Particle Swarm Optimization (PSO), developed by James Kennedy and Russell Eberhart in 1995, is an optimization algorithm that simulates the predatory behavior of birds. In this algorithm, particles share information to find the optimal value. Multi-objective PSO (Multi-Objective Optimization) is an improvement on single-objective PSO by introducing the concept of Pareto optimality and an external archiving mechanism, extending PSO to simultaneously optimize multiple objective functions. In multi-objective optimization problems, the objective function is typically in vector form, represented as: in, Indicates having n An optimization problem with multiple objectives; For sub-objective functions; A representation of the decision space m 3D decision vector.

[0037] In the multi-objective particle swarm optimization algorithm, particles move around in the search space to find the optimal solution. Their positions and velocities need to be continuously updated. t In the next iteration, particles i The update equation is: in, , This represents the velocity of an individual particle; , Indicates the position of an individual particle; Indicates inertia weight; , These represent the cognitive coefficient and the social coefficient, respectively. Represents a random number within the range [0, 1]. , These represent the individual optimality and the global optimality of the particle, respectively.

[0038] This study proposes an integrated strategy for multi-objective particle swarm optimization that combines dynamic parameter adjustment, neighborhood guidance, and adaptive mutation strategies. Dynamic Collaborative Multi-Objective Particle Swarm Optimization (ISMOPSO) is a parameter adaptive optimization method used for remote sensing image enhancement. In the process of population evolution, the... The velocity and position of each particle are updated according to the following equation: In the formula: Represents particles i In iteration t The velocity vector; Indicates inertia weight; These represent the cognitive, social, and neighborhood coefficients, respectively. A random number between [0, 1]; This represents the optimal position of an individual particle; Indicates the globally optimal position; This represents the optimal position in the particle's neighborhood; Represents particlesi In iteration t The position vector.

[0039] In some specific embodiments, when constructing the dynamic cooperative multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the following steps are included: The inertia weight and learning factor are dynamically adjusted using the Sigmoid function, where the dynamic change formula for the inertia weight is: In the formula: , They are 0.9 and 0.4 respectively; For the Sigmoid function; Indicates the current iteration number; , Indicates the maximum number of iterations; The cognitive coefficient C1, social coefficient C2, and neighborhood coefficient C3 are dynamically adjusted using the Sigmoid function, as shown in the following formula: In the formula: Set to 2; ; , Set them to 2.5 and 0.5 respectively; ; Set it to 1.5.

[0040] In some specific embodiments, when constructing the integrated strategy multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the method further includes: constructing a dynamic random neighborhood topology, wherein in each iteration, each particle... neighborhood Generated through random sampling, including: In the formula: Describes the particle in the t-th iteration. The neighborhood set; N Indicates the size of the particle swarm; Indicates the size of the neighborhood; Represents a uniformly random permutation function; in the neighborhood Internally, a dominant relationship is adopted. Choose the solution with the strongest dominance as... If there is no dominant solution, then a non-dominated solution is randomly selected.

[0041] In some specific embodiments, when constructing the integrated strategy multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the method further includes: designing adaptive Gaussian mutation, calculating the normalized difference between the fitness of the current particle and the global optimal solution, dynamically adjusting the mutation probability and mutation amplitude based on this difference, and performing Gaussian perturbation on the particle position, including: the formula for calculating the normalized difference between the fitness of the current particle and the global optimal solution is: in: It is the total number of dimensions of the objective function; It is the dimension of the objective function; It is the fitness vector of the global optimal solution; It is the fitness vector of the current particle; the denominator is... It is the globally optimal k-dimensional absolute value; It is a local minimum. This ensures that attention is paid to dimensions where particle performance is insufficient; The formulas for calculating the probability of mutation and the magnitude of mutation are: in: , representing the basic mutation probability; , represents the adjustment coefficient; , indicating the basic range; , represents the magnification factor; Generate random numbers based on the above parameters. ,like Then a Gaussian mutation is performed, the mutation operation being defined as altering the particle position. Disturbance: in The standard normal random perturbation is standard Gaussian noise that is independent in each dimension.

[0042] In some specific embodiments, the construction of a four-objective fitness function including structural similarity, average gradient, information entropy, and gray-level variance is used to guide the particle to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters, including: The fitness evaluation function formula is: in, This represents an optimization problem with four objectives. Indicators representing structural similarity; Indicates the average gradient; Represents information entropy; Indicates variance; This represents a parameter vector.

[0043] In some specific embodiments, the construction of a four-objective fitness function, including structural similarity, average gradient, information entropy, and gray-level variance, and the use of this function to guide particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters, further includes: The formula for calculating structural similarity is: in, , Images , The average value; , Images , The variance; For image , covariance; , It is a constant; The formula for calculating the average gradient is: in, , The image at the pixel points Horizontal and vertical gradients at the location, and These represent the number of rows and columns of the image, respectively. The formula for calculating information entropy is: in, grayscale The probability of occurrence; The gray level of the image; The formula for calculating the variance of gray levels is: in, For the image at the pixel point The grayscale value at that location; This represents the average grayscale value of the image.

[0044] It should be understood that in multi-objective particle swarm optimization, inertia weight and learning factor are crucial control parameters. Inertia weight determines the impact of a particle's current velocity on its next velocity, while also considering the maintenance of the Pareto front and the diversity of the solution set. The learning factor adjusts the particle's learning behavior and search direction. Particle swarm optimization algorithms typically include inertia weight. Three control parameters, namely the cognitive coefficient C1 and the social coefficient C2, are used. The neighborhood coefficient C3 is added to balance the interaction between particles and local optima. Adjusting the inertia weight and learning factor can effectively regulate the optimization ability of MOPSO. Fixed or dynamically adjusted parameter values ​​can be chosen based on the characteristics of the optimization process. A weighted strategy PSO based on the Sigmoid function is studied. The four parameters are input into the Sigmoid function with different normalized iteration progresses, and scaled using empirical upper and lower bounds to form increasing or decreasing time-varying curves. That is, the Sigmoid function is used to smoothly transition each parameter from its initial value to its final value, enhancing the global search capability in the early stages and focusing on local search in the later stages by outputting a tendency towards the reverse interval, thus improving convergence speed and overall optimization performance. The formula for the dynamic change of the inertia weight with the number of iterations is as follows: In the formula: , They are 0.9 and 0.4 respectively; For the Sigmoid function; Indicates the current iteration number; Normalize the number of iterations to the [0,1] interval to control the curve steepness and center position; This indicates the maximum number of iterations.

[0045] Similarly, the cognitive coefficient C1, social coefficient C2, and neighborhood coefficient C3 are dynamically adjusted using the Sigmoid function, as shown in the following formula: In the formula: Set to 2; ; , Set them to 2.5 and 0.5 respectively; ; Set it to 1.5.

[0046] Figure 4 This paper presents a comparison of the dynamic characteristics of the proposed Sigmoid function adaptive method with three typical dynamic parameter adjustment strategies (linear dynamic parameter adjustment, chaotic dynamic parameter adjustment, and exponentially decaying dynamic parameter adjustment). Figure 4(a) It can be seen that the values ​​of each parameter are adaptively and dynamically adjusted with the number of iterations, and the high inertia weights in the early stage are... High cognitive coefficient C1 ensures global particle search; mid-term inertial weighting. The gradual transition of cognitive factor C1 and the rise of social coefficient C2 enhance the influence between groups; later inertia weights The cognitive factor C1 decreases while the social coefficient C2 reaches its maximum value, promoting convergence and enabling fine-tuning in local areas. From the perspective of multi-target image enhancement, the time-varying parameter adjustment method based on the sigmoid function achieves a smooth transition between the early global search and the later local convergence, and its dynamic change trend is superior to linear, chaotic, and exponential time-varying methods.

[0047] In multi-objective particle swarm optimization, each particle tracks not only its historical best (pbest) and global best (gbest) during iteration, but also the best solutions (lbest) of other particles in its neighborhood. Traditional local neighborhood topologies are mostly static, remaining unchanged throughout the search process. This lack of information exchange between particles can lead to problems such as getting stuck in local optima and premature convergence. To address these issues, research has broken through the limitations of static neighborhoods, proposing Pareto-dominated dynamic stochastic neighborhoods (such as...). Figure 5 (As shown). In each iteration, each particle neighborhood Generated through random sampling, specifically implemented as follows: In the formula: Describes the particle in the t-th iteration. The neighborhood set; N Indicates the size of the particle swarm; Indicates the size of the neighborhood; This represents a uniformly random permutation function. In the neighborhood... Internally, a dominant relationship is adopted. Choose the solution with the strongest dominance as... If there is no dominant solution, then a non-dominated solution is randomly selected.

[0048] Figure 5 This demonstrates the update process of the dynamic random neighborhood topology: Initialization. The particle swarm is randomly distributed in the search space ( Figure 5 a) Neighborhood selection. Each particle (red) randomly selects 5 neighbors (connected by blue lines) to form a temporary neighborhood. Figure 5 b) Neighborhood optimal update. The optimal solution within the neighborhood is selected as the lbest (green) to guide particle movement. Figure 5 c).

[0049] Iterative update. In the next iteration, a new neighborhood is randomly selected, and the above process is repeated.

[0050] Traditional MOPSO is prone to premature convergence, mainly because particles over-rely on global and individual optimal information, leading to a decline in population diversity. Therefore, when the population evolves to a certain extent, mutation operations are performed to increase its diversity, prompting mutated particles to explore new regions and potentially discover new optimal solutions in subsequent evolution. Commonly used mutation operators include Gaussian mutation (GM) and Cauchy mutation (CM). This study proposes an adaptive mutation method that dynamically adjusts the mutation probability and variational time, and combines Gaussian mutation to perform the mutation operation.

[0051] For each objective function, calculate the fitness of the current particle. With the global optimal solution To avoid the influence of differences in the dimensions of different objectives, and to normalize the differences in fitness: In the formula: It is the dimension of the objective function; It is the fitness vector of the global optimal solution; It is the fitness vector of the current particle; the denominator is... It is the globally optimal k-dimensional absolute value (normalized dimensions). It is the minimum value (to prevent the denominator from being zero). This ensures that the dimensions of the particles being monitored are not performing well (if a certain dimension is already better than the global optimum, the difference is set to 0).

[0052] based on Calculate adaptive mutation probability With the magnitude of variation When a particle's fitness deviates significantly from the global optimum, the mutation probability and magnitude are increased to enhance global search capabilities; when a particle approaches the optimal solution, mutation is reduced to maintain solution stability. The mutation probability and magnitude are dynamically adjusted based on the particle's fitness. In the formula: , representing the basic mutation probability; , represents the adjustment coefficient; , indicating the basic range; , represents the magnification factor.

[0053] Generate random numbers based on the above parameters. ,like Then a Gaussian mutation is performed, the mutation operation being defined as altering the particle position. Disturbance: in The standard normal random perturbation is standard Gaussian noise that is independent in each dimension.

[0054] The study used six evaluation indicators: standard deviation, information entropy (Formula 16), visual information fidelity, average gray level (brightness), spatial frequency, and average gradient (Formula 15) to comprehensively evaluate the remote sensing image enhancement results from aspects such as image contrast, content richness, image visual quality, and texture detail.

[0055] Standard deviation (StdDev) measures the dispersion of gray values ​​in an image and reflects the image's contrast. Its definition is as follows: in, StdDev is the average grayscale value of the image. A higher StdDev value indicates a higher image contrast.

[0056] Visual Information Fidelity (VIF), based on information theory, evaluates the degree to which visual information is preserved in an image. It measures the extent to which visual information is retained by quantifying the mutual information between the original and enhanced images. A higher VIF value indicates better visual information fidelity.

[0057] The average gray level (mean) reflects the overall brightness level of an image and is defined as follows: Spatial frequency (SF) measures the frequency of grayscale changes in an image, reflecting the image's sharpness and detail. Its definition is as follows: Here, RF and CF are the row frequency and column frequency, respectively, calculated as the root mean square of the gray-level differences in the row and column directions of the image. A higher SF value indicates richer image details.

[0058] The results of the enhancement of 6 sets of remote sensing images are as follows Figure 6 As shown, the processing effects of the original remote sensing images (a1–a6) under various enhancement methods are demonstrated, including ISMOPSO-HC (b1–b6), MOPSO-HC (c1–c6), HE (d1–d6), WT (e1–e6), SSR (f1–f6), LAGC (g1–g6), and HF (h1–h6).

[0059] From a visual perspective, ISMOPSO-HC ( Figure 6 b1–b6) demonstrated good terrain texture enhancement effects in all test images while maintaining overall image brightness balance. This method highlights subtle features and excels in brightness adaptation and local contrast control, effectively improving the overall visual perception quality of the images. However, the overall image enhancement results exhibit some color distortion. (Images) Figure 6 The area selected by the box (b1–b6) is a typical mountainous area, showing a high degree of topographical recognition and structural integrity. The ridge lines are clear, the gullies are continuous, the light and shadow transitions are natural, and there is no obvious over-enhancement.

[0060] MOPSO-HC ( Figure 6 c1–c6 also showed good enhancement effects. In terms of detail, the ridge and gully structures were relatively clear, and the overall contrast was moderate. However, compared with ISMOPSO-HC, the overall color of the image was slightly darker. In (c3) and (c4), the image details in the low-light areas were slightly lost, and some gully textures were not fully displayed.

[0061] Comparison of HE methods ( Figure 6 While images d1–d6 significantly improved overall brightness and grayscale contrast, their global mapping characteristics prevented them from effectively adapting to terrain structures with significant local brightness variations in remote sensing images, resulting in oversaturation of image edge information. For example, image (d3) exhibited excessively strong local contrast, leading to overly sharp edges, distorted textures, insufficient enhancement of terrain ridges and gullies, poor enhancement in low-light areas, and even a "halo" phenomenon in some areas.

[0062] WT method ( Figure 6 The images (e1–e6) perform well in terms of detail, highlighting certain structural features of ridges and valleys. They have a certain advantage in improving the edge sharpness of the images. However, in some images (e4, e5, e6), the overall color shows saturation imbalance and high-frequency false color phenomenon. Furthermore, the overall enhancement results show little improvement in brightness and contrast, and details in low-light areas are not obvious enough, affecting the authenticity of remote sensing image interpretation.

[0063] The SSR method has certain advantages in restoring information in the shadow areas of remote sensing images. The overall brightness of the image f1-f6 is improved and the dark areas are significantly brightened, which improves visual visibility. However, blurring occurs at the transition between the bright and shadow areas. The enhancement of details such as ridges and gullies in remote sensing images is not obvious, and the sense of structure is poor.

[0064] The LAGC method achieved good texture enhancement in some images (such as g3), but the overall contrast improvement was limited, the terrain texture was not obvious, it was difficult to highlight the mountain features, and there was local saturation, indicating that the enhancement ability was slightly insufficient. In images g2 and g4, it can be seen that the detail representation of mountain gullies is limited, and the overall contrast of the enhancement result is low.

[0065] from Figure 6 As can be seen from h1-h6, the HF method has a certain corrective effect when processing remote sensing images with shadows and uneven lighting. It can preserve a certain degree of large-scale brightness structure, and the overall image color is relatively close to the original image. However, the improvement in overall brightness and texture details is limited, and it fails to fully reflect the terrain complexity of the remote sensing image.

[0066] In summary, the ISMOPSO-HC method significantly improves the texture clarity and contrast of typical mountainous terrain features (such as ridges and gullies) in remote sensing images, while effectively suppressing over-enhancement and halo artifacts during image enhancement. Although some color distortion exists, the overall enhancement effect on terrain features in low to medium illumination is good, making it an effective means of enhancing remote sensing images with complex terrain. This also confirms that the HF-CLAHE parameter optimization driven by ISMOPSO achieves Pareto optimality in a multi-objective trade-off.

[0067] To further evaluate the preservation of topographic structure information in remote sensing images by different enhancement methods, normalized autocorrelation function (NAC) analysis was used to analyze image structure characteristics, and structural similarity was evaluated within a two-dimensional offset range for images before and after enhancement. By calculating the NAC coefficients of the images under different horizontal and vertical translations, the texture repeatability and spatial structure consistency of the enhanced images were reflected, thereby quantifying the degree to which the enhancement algorithms preserved and enhanced the original structural information.

[0068] Figure 7 This paper presents 3D surface plots of the structural autocorrelation of six remote sensing images under seven enhancement methods. In the 3D surface plots, the X and Y axes represent the horizontal and vertical translation distances of the image, respectively, and the Z axis represents the autocorrelation coefficient of the image under the corresponding translation. If the autocorrelation plot of the enhanced image shows a sharp, symmetrical central peak near the origin, it indicates that the image retains good spatial structural continuity and directional consistency. If the peak diffuses, the shape flattens, or there is a shift, it means that the enhancement process may have resulted in loss of structural information, blurred texture, or local deformation.

[0069] Figure 7 a1–a6 are autocorrelation surface plots of the original remote sensing images. It can be observed that these images all exhibit symmetrical and clear central peaks, indicating that the original images possessed a certain degree of topographic structural continuity and texture repeatability before processing. Compared to the original images, the ISMOPSO-HC method shows better performance in structural autocorrelation analysis (…). Figure 7The image (b1–b6) exhibits a broader peak-to-base pattern, while the peak values ​​remain sharp and the center is prominent. Its autocorrelation distribution maintains high correlation values ​​within the offset range, indicating significant structural repeatability in multiple directions. This phenomenon reflects both the algorithm's overall ability to enhance terrain texture and the better preservation and extension of detailed structures in space. Quantitatively, the autocorrelation coefficient of our method remains above 0.6, significantly higher than other methods. Compared to traditional homomorphic filtering or WT, it has a higher autocorrelation peak, a smoother decay curve, and significantly enhanced spatial consistency.

[0070] Further comparison with the MOPSO-HC method reveals that the ISMOPSO-HC method, through its improved multi-objective particle swarm search strategy, can selectively preserve local structures that significantly influence autocorrelation, avoiding overemphasis on low-frequency regions during particle optimization. Analysis of the autocorrelation plot shows that the MOPSO-HC method ( Figure 7 While the C1-C6 graphs improved structural consistency to some extent, the peaks in some graphs (C3 and C5) were slightly blurred and the edge expansion areas were wide, indicating that the detail contrast was not improved enough during the enhancement process. This suggests that the MOPSO algorithm may still have search biases in local areas.

[0071] Conventional enhancement algorithms such as HE, SSR, and WT exhibit concentrated peaks and rapid decay on the flanks in the structure autocorrelation plot. The central peaks in d5 and f5 spread outwards, indicating that the enhancement process introduces significant global gray-level redistribution, disrupting the repeatability of the original structure. Under the LAGC method, peaks in g3 and g6 deviate from the center, and even g5 shows significant diffusion, suggesting that while LAGC can improve local contrast, it may introduce pseudo-structural information, resulting in slightly lower structural stability compared to ISMOPSO-HC.

[0072] In summary, ISMOPSO-HC demonstrates significant advantages in structural autocorrelation analysis after image enhancement. The main peak of the autocorrelation surface in the enhanced image is more prominent, indicating effective enhancement of local structural details; the wider main peak region reflects improved overall structural consistency; the autocorrelation distribution exhibits clear directionality, reflecting the enhancement and preservation of texture orientation; and the expanded autocorrelation response range indicates a significant improvement in image texture repeatability, consistent with subjective evaluation results. These characteristics collectively demonstrate the effectiveness of the ISMOPSO-HC method in enhancing the structural information of remote sensing images, providing fundamental image support for practical applications (geological hazard identification, landslide boundary extraction, and forest resource analysis).

[0073] To comprehensively evaluate the performance of different remote sensing image enhancement methods, this study used six commonly used image quality evaluation indicators for quantitative analysis, including image standard deviation (StdDev), information entropy (Entropy), visual information fidelity (VIF), image mean gray level (Mean), spatial frequency (Spatial Frequency), and average gradient (AvgGradient) (Tables 1-6): Table 2 Comparison of image standard deviations for different algorithms

[0074] Table 2 shows the image standard deviation results. Among the six remote sensing images, ISMOPSO-HC had the highest standard deviation (mean 0.2528±0.0114), which was significantly higher than other comparison methods. Compared with the second-best HE (0.2125±0.0059), it improved by 18.96%. The image enhanced by the LAGC method had the lowest standard deviation. The results indicate that ISMOPSO-HC can effectively improve the contrast of remote sensing images.

[0075] Table 3 Comparison of Image Information Entropy Results of Different Algorithms

[0076] Table 3 shows the image information entropy results. The entropy value of the enhanced image by the ISMOPSO-HC method (mean 7.8410 ± 0.0722) is higher than that of other comparison methods, indicating that its enhanced image has the best information richness. Especially in high-frequency detail scenes (such as a2), the entropy value reaches 7.9086, which is 19.14% higher than WT (6.6382), indicating that it can effectively improve the expression of texture complexity.

[0077] Table 4 Comparison of Image Visual Information Fidelity of Different Algorithms

[0078] As shown in Table 4 regarding the fidelity of visual information in the images, the ISMOPSO-HC method outperforms other methods in terms of VIF (mean 5.3648 ± 0.7804) across all six images, achieving a 42.25% improvement over the second-best method, MOPSO (3.7713 ± 0.8673). This result quantitatively verifies that the ISMOPSO-HC method enhances details while preserving the original structural information to the greatest extent possible. Conventional methods (such as SSR) suffer significant loss of structural information due to brightness normalization.

[0079] Table 5 Comparison of average grayscale values ​​of images using different algorithms

[0080] Table 5 shows that the average brightness of the ISMOPSO-HC method is stable at 0.4289±0.0304, which is close to the ideal intermediate gray level (0.5). The enhancement process avoids excessive brightness enhancement or suppression. Compared with WT (0.1468), it is more in line with the natural brightness of the image, indicating that the automatic parameter adjustment of the ISMOPSO-HC method can achieve better brightness balance.

[0081] Table 6 Comparison of image spatial frequency results for different algorithms

[0082] In Table 6, the ISMOPSO-HC method performs best in terms of spatial frequency (0.1626±0.0088), which is significantly higher than other methods (the second best is MOPSO-HC: 0.1315±0.0159), indicating that it has a significant advantage in enhancing terrain details and is more suitable for high-precision terrain information extraction and analysis.

[0083] Table 7 Comparison of image average gradient results for different algorithms

[0084] As shown in Table 7, the average gradient of ISMOPSO-HC reached 0.7633±0.0560, which is 24.95% higher than that of the suboptimal MOPSO-HC (0.6109±0.0839). In the a3 image, the gradient value (0.8448) is 2.57 times that of SSR (0.3292). This result is consistent with the subjective evaluation. The ISMOPSO-HC enhancement strategy significantly improves the image sharpness and edge representation, especially in high-frequency structures such as gullies and slopes.

[0085] In summary, the ISMOPSO-HC method outperforms conventional enhancement methods and optimization algorithms (MOPSO) in all six quantitative metrics, particularly in image detail richness (entropy), structure fidelity (VIF), and texture representation (spatial frequency and gradient). The metric evaluation results are highly consistent with subjective evaluations and structural autocorrelation analysis, further validating the adaptability and stability of ISMOPSO-HC under different terrain conditions.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A homomorphic filter-CLAHE remote sensing image enhancement method integrated with policy multi-objective particle swarm optimization, characterized in that, Includes the following steps: Acquire remote sensing image data and preprocess it, including radiometric calibration, atmospheric correction and spatial cropping; The preprocessed remote sensing image is converted to the HSV color space, and the V channel component is extracted. A frequency-spatial hybrid enhancement framework combining homomorphic filtering and contrast-limited adaptive histogram equalization is constructed, and this framework is used to enhance the V channel components. An integrated strategy multi-objective particle swarm optimization algorithm is constructed to optimize the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework. A four-objective fitness function, including structural similarity, average gradient, information entropy and gray variance, is constructed. This function guides the particle to search in the direction of simultaneously optimizing multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters. When constructing the dynamic cooperative multi-objective particle swarm optimization algorithm to optimize the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm includes: The inertia weight and learning factor are dynamically adjusted using the Sigmoid function, where the dynamic change formula for the inertia weight is: In the formula: , are respectively 0.9, 0.4; is a Sigmoid function; denotes the current iteration number; , denotes the maximum iteration number; The cognitive coefficient C1, social coefficient C2, and neighborhood coefficient C3 are dynamically adjusted using the Sigmoid function, as shown in the following formula: In the formula: Set to 2; ; , Set them to 2.5 and 0.5 respectively; ; Set it to 1.5; When constructing the integrated strategy multi-objective particle swarm optimization algorithm for optimizing the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm further includes: constructing a dynamic random neighborhood topology, where each particle... neighborhood Generated through random sampling, including: In the formula: Describes the particle in the t-th iteration. The neighborhood set; N Indicates the size of the particle swarm; Indicates the size of the neighborhood; Represents a uniformly random permutation function; in the neighborhood Internally, a dominant relationship is adopted. Choose the solution with the strongest dominance as... If there is no dominant solution, then a non-dominated solution is randomly selected. When constructing the integrated strategy multi-objective particle swarm optimization algorithm to optimize the parameters of homomorphic filtering and contrast-limited adaptive histogram equalization in the frequency-spatial hybrid enhancement framework, the algorithm further includes: designing adaptive Gaussian mutation, calculating the normalized difference between the fitness of the current particle and the global optimum, dynamically adjusting the mutation probability and mutation amplitude based on this difference, and performing Gaussian perturbation on the particle position, including: the formula for calculating the normalized difference between the fitness of the current particle and the global optimum is: in: It is the total number of dimensions of the objective function; It is the dimension of the objective function; It is the fitness vector of the global optimal solution; It is the fitness vector of the current particle; the denominator is... It is the globally optimal k-dimensional absolute value; It is a local minimum. This ensures that attention is paid to dimensions where particle performance is insufficient; The formulas for calculating the probability of mutation and the magnitude of mutation are: in: , representing the basic mutation probability; , represents the adjustment coefficient; , indicating the basic range; , represents the magnification factor; Generate random numbers based on the above parameters. ,like Then a Gaussian mutation is performed, the mutation operation being defined as altering the particle position. Disturbance: in The standard normal random perturbation is standard Gaussian noise that is independent in each dimension.

2. The method for enhancing remote sensing images using homomorphic filtering-CLAHE remote sensing image enhancement based on integrated strategy multi-objective particle swarm optimization according to claim 1, characterized in that, When constructing the frequency-spatial hybrid enhancement framework of homomorphic filtering and contrast-limited adaptive histogram equalization, and using this framework to enhance the V channel component, the following steps are included: image Decomposed into low-frequency irradiation components and high-frequency reflection components : Perform a logarithmic transformation on the image to separate it into low-frequency illumination. With high frequency reflection : Performing a Fourier transform yields the frequency domain expression: in, and These represent the frequency domain components of the illumination and reflection components, respectively; In the frequency domain space Perform Gaussian filtering : Reconstructed and enhanced spatial image: The V-channel components are filtered using an optimized Gaussian homomorphic filter, including: The transfer function of the optimized Gaussian homomorphic filter is: ； in, , These represent high-frequency gain and low-frequency gain, respectively. and Separately control high-frequency enhancement and low-frequency attenuation; This represents the distance from a point in the frequency domain to the center. Image size, The cutoff frequency, This is the steepness parameter of the transition zone.

3. The method for enhancing remote sensing images using homomorphic filtering-CLAHE remote sensing image enhancement based on integrated strategy multi-objective particle swarm optimization according to claim 2, characterized in that, When using the frequency-spatial hybrid enhancement framework for constructing homomorphic filtering and contrast-limited adaptive histogram equalization to enhance the V channel component, the following further steps are included: A contrast-limited adaptive histogram equalization algorithm is applied to the V channel component after homomorphic filtering. This algorithm includes dividing the image into sub-blocks, truncating and redistributing the histogram of each sub-block, generating a normalized cumulative distribution function and mapping grayscale values, and interpolating the mapping functions of adjacent sub-blocks. Image Divided into Each sub-block histogram Defined as: in, For gray levels, It is the Dirac function; Introduce a contrast limiting threshold Truncate the histogram for each sub-block: Where M is the number of pixels in the sub-block; After truncation, the histogram maintains the total number of pixels through linear redistribution: Generate normalized CDF And map grayscale values: Interpolate the mapping function of adjacent sub-blocks to output pixel values. for: Among them, weight Determined by pixel position.

4. The method for enhancing remote sensing images using homomorphic filtering-CLAHE based on integrated strategy multi-objective particle swarm optimization according to claim 3, characterized in that, The four-objective fitness function, including structural similarity, average gradient, information entropy, and gray-level variance, is constructed. This function guides the particle to search in a direction that simultaneously optimizes multiple key image quality indicators, achieving the optimal selection of remote sensing image enhancement parameters, including: The fitness evaluation function formula is: in, This represents an optimization problem with four objectives. Indicators representing structural similarity; Indicates the average gradient; Represents information entropy; Indicates variance; This represents a parameter vector.

5. The method for enhancing remote sensing images using homomorphic filtering-CLAHE remote sensing image enhancement based on integrated strategy multi-objective particle swarm optimization according to claim 4, characterized in that, The construction of a four-objective fitness function, including structural similarity, average gradient, information entropy, and gray-level variance, guides particles to search in a direction that simultaneously optimizes multiple key image quality indicators, thereby achieving the optimal selection of remote sensing image enhancement parameters. This also includes: The formula for calculating structural similarity is: in, , Images , The average value; , Images , The variance; For image , covariance; , It is a constant; The formula for calculating the average gradient is: in, , The image at the pixel points Horizontal and vertical gradients at the location, and These represent the number of rows and columns of the image, respectively. The formula for calculating information entropy is: in, grayscale The probability of occurrence; The gray level of the image; The formula for calculating the variance of gray levels is: in, For the image at the pixel point The grayscale value at that location; This represents the average grayscale value of the image.

Images