Corn sprout image segmentation method based on elite adaptive rime algorithm

Abstract

A corn sprout image segmentation method based on an elite adaptive rime algorithm. This invention relates to the technical fields of agricultural seed detection and image processing, and more specifically to the technical field of corn sprout image segmentation based on an elite adaptive rime algorithm. The present invention introduces a dual adaptive weight mechanism and an elite reselection strategy into the rime optimization algorithm to enhance the algorithm's convergence capability, and combines Kapur entropy to perform multi-threshold segmentation on corn kernel sprout images. This effectively improves the accuracy of image segmentation. The method comprises the following steps: obtaining a corn sprout image dataset; plotting a two-dimensional histogram and inputting it into a Kapur entropy function to obtain an objective function fobj; initializing an elite solution module; adding a dual adaptive weight mechanism; improving the soft rime strategy and the hard rime strategy; updating the elite solution after the rime search strategy module; and inputting the objective function fobj into the improved rime algorithm to obtain an optimal threshold.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural seed detection and image processing, and in particular to the technical field of corn germination image segmentation based on an elite adaptive rime algorithm. Background Art

[0002] In agricultural production, corn, one of the world's three major staple crops, has a significant impact on global food security, with its yield and quality. Accurate segmentation of corn kernel germination images can help agricultural experts more accurately assess seed quality and predict germination rates, playing a crucial role in seed testing, variety selection, and agricultural planting decisions. Failure to accurately and timely analyze seed germination can lead to inaccurate planting decisions and inefficient variety selection, significantly impacting the agricultural economy. Accurate segmentation of crop seed images is an essential requirement for scientific management and a key factor in improving agricultural productivity.

[0003] Currently, the process of analyzing seed germination images using machine vision typically consists of several distinct stages. First, the image undergoes preprocessing to reduce noise. Next, the image is segmented to isolate specific regions within the target image. Finally, the segmented image undergoes feature extraction and analysis. Image segmentation is a key component of this entire process, as it significantly impacts subsequent image analysis.

[0004] In recent years, researchers have proposed several effective image segmentation algorithms. Among all image segmentation algorithms, threshold segmentation is computationally simple and robust. Threshold segmentation first uses a grayscale histogram to perform image thresholding to determine the threshold that distinguishes different classes within a given image. When a single threshold is used to separate two classes, a secondary threshold is generated, also known as image binarization.

[0005] In contrast, multi-level thresholding (MTH) classifies an image into multiple categories, exceeding the number of categories. At the same time, as the combination of more thresholds increases the complexity of the search process, the accuracy of image segmentation decreases. Therefore, compared with multi-level thresholding, two-level thresholding is simpler to implement. Threshold-based segmentation methods can be divided into parametric and non-parametric methods. Parametric techniques use probability and density functions to define the attributes of all categories, but they require a lot of computing resources. In contrast, non-parametric methods use metrics such as variance, entropy, and error rate to assess differences between groups, reducing the waste of computing resources.

[0006] Traditional multi-threshold segmentation algorithms (MTIS) use exhaustive enumeration to determine the optimal threshold, resulting in exponentially increasing computational complexity with the number of thresholds and low computational efficiency. To address this issue, researchers have used swarm intelligence algorithms to calculate the optimal threshold, alleviating this computational inefficiency to a certain extent. Common optimization algorithms include particle swarm optimization (PSO), genetic algorithms (GA), and gray wolf optimization (GWO). However, swarm intelligence algorithms are limited by the No Free Lunch (NEL) principle, which prevents them from achieving good results in all areas. In the field of agricultural seed detection, swarm intelligence algorithms, due to the limitations of their search mechanism, are prone to premature convergence to local optima and are unable to effectively integrate global and local searches, resulting in low accuracy in agricultural seed status detection and identification. Summary of the Invention

[0007] To address the problem of existing swarm intelligence algorithms prematurely converging and falling into local optima, this paper proposes a corn sprout image segmentation method based on an elite adaptive rime algorithm. This method incorporates a dual adaptive weighting mechanism and an elite reselection strategy into the rime optimization algorithm to enhance its convergence. Furthermore, Kapur entropy is used to perform multi-threshold segmentation on corn kernel sprout images, effectively improving image segmentation accuracy.

[0008] The method comprises the following steps:

[0009] S1. Obtain a corn sprout image dataset and perform preprocessing to obtain a grayscale image and a non-local mean image;

[0010] S2, draw a two-dimensional histogram based on the grayscale image and the non-local mean image, and input it into the Kapur entropy function to obtain the objective function fobj;

[0011] S3. Construct and improve the rime algorithm to obtain the improved rime algorithm:

[0012] S31. Add an initialization elite solution module after the initialization module;

[0013] The initialization elite solution module specifically comprises the following steps: sorting the initial fitness values ​​of the rime population in ascending order, and selecting the first three rime particles according to the sorting result as the initial elite solution: 、 and ;

[0014] S32. Add a dual adaptive weight mechanism between the initialization elite solution module and the rime search strategy module;

[0015] Furthermore, the dual adaptive weight mechanism is specifically as follows:

[0016] S321, calculate global search weight and local search weight : , ,in, Indicates the current evaluation number, Indicates the maximum number of evaluations;

[0017] S322, according to Calculate the parameters used to control the search step size of the improved soft rime strategy , specifically: ,in, express A random number between Represents pi and the number of current function evaluations The product of represents the rounding function, Indicates soft rime parameters;

[0018] S323, according to Calculate the trigger probability parameters of the improved strategy for soft rime respectively and Hard Rime Improved Strategy Triggering Probability Parameters ,in, , ,in, represents the square root function, Represents the normalized value of the fitness value of all rime particles, Indicates the The fitness value of each rime particle;

[0019] S33. Using the elite reselection strategy to improve the soft rime strategy and the hard rime strategy, obtaining the soft rime improvement strategy and the hard rime improvement strategy;

[0020] The soft rime improvement strategy is specifically as follows: ,in, Indicates the In terms of dimension, The optimal solution of the improved soft rime strategy for rime particles is searched. Indicates the In the dimension, the initial elite solution is selected according to the probability, where the selection 、 and The probability ratio is 6:3:1;

[0021] The hard rime improvement strategy is specifically as follows: ,in, Indicates the In terms of dimension, The improved hard rime strategy of rime particles searches for the optimal solution;

[0022] S34, update the elite solution after the rime search strategy module;

[0023] S4, input the objective function fobj into the rime improved algorithm to obtain the optimal threshold;

[0024] S5. Segment the grayscale image obtained in step S1 according to the optimal threshold value to obtain a segmented image.

[0025] Furthermore, the preprocessing specifically includes: sequentially performing grayscale processing and non-local mean filter processing.

[0026] Furthermore, the updating of the elite solution after the rime search strategy module is specifically as follows: sorting the individual fitness values ​​of the population updated by the rime search strategy module in ascending order, updating the initial elite solution according to the sorting result, and obtaining an updated elite solution: 、 and .

[0027] Furthermore, the objective function fobj is input into the rime improved algorithm to obtain the optimal threshold, specifically:

[0028] S41, initialize the rime population, , the upper limit of the search dimension , the lower limit of the search dimension and fobj;

[0029] S42. Obtain the initial elite solution: 、 and ;

[0030] S43, using the dual adaptive weight mechanism to calculate respectively: 、 and ;

[0031] S44, when When implementing the soft rime improvement strategy, express Otherwise, execute step S45;

[0032] S45, when When implementing the hard rime improvement strategy, express Otherwise, execute step S46;

[0033] S46. Get the updated elite solution: 、 and ;

[0034] S47, when When , the elite solution obtained in step S46 is used as the initial elite solution, and steps S43 to S47 are executed in a loop;

[0035] Otherwise, the evaluation ends and the optimal threshold is output.

[0036] The beneficial effects of the method of the present invention are:

[0037] (1) The improved rime algorithm provided by the present invention combines the elite reselection strategy and the dual adaptive weight mechanism, which can enhance the convergence ability of the algorithm and provide a more favorable area (optimal threshold) for subsequent image segmentation.

[0038] (2) The method described in the present invention uses Kapur entropy to process the two-dimensional histogram of grayscale images and non-local mean images, and uses the output as the input of the rime improvement algorithm. Together with the rime improvement algorithm, it acts on the multi-domain value segmentation of corn kernel germination images, effectively improving the accuracy of image segmentation and the detection and recognition accuracy of corn kernel germination status, providing effective technical support for seed quality assessment and agricultural planting decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the improved algorithm for rime according to the present invention;

[0040] Figure 2 This is a schematic diagram of the threshold segmentation of corn sprout images according to the present invention;

[0041] Figure 3 The original corn sprout image of the present invention;

[0042] Figure 4 The image after segmentation when the threshold value is 6 according to the present invention;

[0043] Figure 5 This is the color mapping image described in the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] Example 1

[0046] This embodiment provides a corn sprout image segmentation method based on the elite adaptive rime algorithm, the method comprising the following steps:

[0047] like Figure 2As shown in FIG, a corn germination image dataset is obtained. First, the corn germination image dataset is gray-scale processed to create a gray-scale image. Then, based on the gray-scale image, a non-local mean (NLM) filtering process is performed to obtain an NLM image (non-local mean image).

[0048] Draw a two-dimensional histogram based on the grayscale image and the non-local mean image, extract the pixel values ​​embedded in the histogram, and input it into the Kapur entropy function to obtain the objective function fobj:

[0049] Kapur entropy is an image segmentation method based on information entropy theory. It determines the optimal threshold by maximizing the sum of entropy between segmented regions. Based on quadrant analysis of a two-dimensional histogram, the method described in this paper constructs a continuous K interval divided along the main diagonal to extract image features. The method also incorporates Kapur entropy theory to establish a feature modeling method for grayscale distribution. The mathematical expression of the modeling method is:

[0050]

[0051] in, represents the total entropy function that needs to be maximized, and Respectively represent the grayscale and gradient direction extracted from the two-dimensional histogram Segmentation threshold points, divide the two-dimensional histogram into different areas; Indicates that the pixel has both grayscale values a and gradient values b The joint probability density of Represents the sum of the probabilities of all pixels in each segmented area; Represents the number of gray levels of the image.

[0052] The method of the present invention can convert the optimal segmentation threshold into a function by mathematical modeling method. The optimal segmentation threshold can be obtained by solving the following objective function: ,in, represents the optimal segmentation threshold, Represents the maximum value calculation function. In this embodiment, Represents the improved algorithm for rime. In this embodiment It represents the input objective function fobj of the improved rime algorithm, which is obtained by embedding the two-dimensional histogram into the Kapur entropy function.

[0053] This paper improves upon the Rime Emulsion Optimization (RIME) algorithm, a relatively recent and highly successful swarm intelligence algorithm. Inspired by the motion of soft frost particles, this algorithm simulates their motion behavior to perform search operations and incorporates the cross-interaction mechanism of hard frost formation to enhance global optimization capabilities.

[0054] The rime search strategy module in the rime optimization algorithm includes: soft rime strategy and hard rime strategy;

[0055] In a breeze, the growth of soft rime is highly random, allowing rime particles to freely cover most of the surface of the object, but growing slowly in the same direction. Inspired by the growth of soft rime, we exploit the strong randomness of rime particles and their ability to cover a large area on an object to develop a soft rime strategy.

[0056]

[0057]

[0058]

[0059]

[0060] In the initialization phase, the rime particles first complete the status update according to the soft rime strategy. The position after the soft rime strategy update is recorded as ,in and Respectively represent The first dimension Rime particles. Indicates the The particle with the best performance in each dimension. Indicates a range in A random number between , used to control the direction of particle update. It represents the parameter that describes the randomness of the adhesion between particles and controls the spacing between particles. Its value is taken from , Indicates the current number of iterations. When the number of iterations reaches the preset maximum value When , the iteration ends. represents the parameter used to control the periodic oscillation behavior in the rime optimization algorithm, represents the parameters that simulate the evolutionary influence of the external environment, Indicates rounding operation, Indicates the parameters for adjusting the function running stage, usually set to 5. and Respectively represent The first dimension The search upper and lower limits of rime particles, represents the adhesion coefficient that increases with the iteration process, characterizing the adsorption strength between particles; and the random variable (express Random numbers in the interval) and The coupling effect determines the aggregation and diffusion behavior of rime particles.

[0061] In strong winds, the formation of hard rime is simpler and more regular than that of soft rime, resulting in puncture. Inspired by the frost puncture phenomenon, the RIME algorithm proposes a hard rime strategy. The replacement relationship between particles is as follows:

[0062]

[0063] in Indicates the optimal rime particle position after the hard rime strategy is updated. Indicates the The best rime particles in terms of dimension. is the normalized fitness value of all rime particles, express Random value in the interval.

[0064] The method of the present invention improves the rime algorithm as follows to obtain an improved rime algorithm:

[0065] S1. Add the initialization elite solution module after the initialization module;

[0066] S2. Add a dual adaptive weight mechanism between the initialization elite solution module and the rime search strategy module;

[0067] S3. Use the elite reselection strategy to improve the soft rime strategy and the hard rime strategy to obtain the soft rime improvement strategy and the hard rime improvement strategy;

[0068] S4. Update the elite solution after the rime search strategy module.

[0069] like Figure 1 As shown in Figure 2, the improved rime algorithm includes the following steps:

[0070] (1) Population initialization and maximum number of initialization evaluations , the upper limit of the search dimension , the lower limit of the search dimension and fobj.

[0071] The rime population is composed of Rime Agent Each rime agent consists of Rime particles Composition, of which Indicates the Rime particles, Indicates the First, initialize each rime particle, then evaluate the population and calculate the fitness value of each rime particle. Represents the fitness value of each rime particle, and arranges the rime particles in ascending order according to their fitness values. The fitness of rime particles is arranged in ascending order Specifically:

[0072]

[0073]

[0074] (2) Obtaining the initial elite solution: Sort the initial fitness values ​​of the rime population in ascending order. Based on the elite reselection strategy, select the first three rime particles according to the sorting results as the initial elite solution: 、 and .

[0075] (3) In order to achieve a dynamic balance between exploration and development at different stages of the improved rime algorithm, a dual adaptive weight mechanism is introduced. This mechanism uses two weight parameters and Controls the strength of global and local searches respectively.

[0076]

[0077]

[0078] in, Indicates the current evaluation number;

[0079] according to Calculate the parameters used to control the search step size of the improved soft rime strategy , specifically: ,in, express A random number between Represents pi and the number of current function evaluations The product of represents the rounding function, Indicates the soft rime parameter, usually set to 5;

[0080] Used to control the search step size of the soft rime improvement strategy. The adjustment of , realizes the smooth transition from the early large step size global exploration to the later small step size fine search. and As the iteration progresses, the disturbance amplitude gradually decreases and the exploration ability of the algorithm gradually decreases.

[0081] according to Calculate the trigger probability parameters of the improved strategy for soft rime respectively and Hard Rime Improved Strategy Triggering Probability Parameters ,in, , ,in, represents the square root function, Represents the normalized value of the fitness value of all rime particles, Indicates the The fitness value of each rime particle.

[0082] Used to control the trigger probability of the soft rime improvement strategy, through Adjustment can achieve the transition from early low-frequency triggering to maintain diversity and later high-frequency triggering to strengthen local development.

[0083] Used to control the probability of triggering the hard rime improvement strategy, in the early stage Small, the probability of triggering the hard rime improvement strategy is low, avoiding premature convergence; in the later stage The larger the value, the higher the probability of triggering the hard rime improvement strategy, which accelerates the final convergence.

[0084] In the original Rime Optimization Algorithm (RIME), both the soft rime strategy and the hard rime strategy only learn from a single optimal solution, Best_rime, which can easily lead to the algorithm converging to a local optimum prematurely. To solve this problem, the elite reselection strategy is introduced. This mechanism maintains three elite solutions ( 、 and ), providing diverse search directions for the algorithm.

[0085] The specific strategies for improving soft rime are: ,in, Indicates the In terms of dimension, Search for the optimal solution for soft rime with rime particles. Indicates the In the dimension, the initial elite solution is selected according to the probability, where the selection 、 and The probability ratio is 6:3:1.

[0086] Alpha represents the optimal position that the population can search, Beta represents the individual with the lowest fitness value in the population other than Alpha, and Delta represents the individual with the lowest fitness value in the population other than Alpha and Beta. The elite reselection strategy divides the population's search behavior into three guiding directions, providing richer evolutionary information through a multi-leader mechanism.

[0087] The specific improvement strategies for hard rime are: ,in, Indicates the In terms of dimension, Search for the optimal solution for hard rime with rime particles.

[0088] Through the hierarchical probability guidance, the main convergence trend towards the optimal solution is guaranteed while the diverse distribution of solutions is maintained.

[0089] like Figure 1 As shown, when When , the soft rime improvement strategy is executed, otherwise, the and The relationship among them, express A random number between

[0090] when When , the hard rime improvement strategy is implemented, otherwise, the new solution is evaluated, where express A random number between

[0091] Evaluating the new solution means calculating the single-modal function, multi-modal function, hybrid function and combination function of the current evaluation result according to the CEC2017 test function standard in each iteration round.

[0092] After evaluating the new solution, perform forward greedy selection to update the initial elite solution (elite reselection): 、 and , that is, sort the updated individual fitness values ​​of the population in the rime search strategy module (soft rime improvement strategy and hard rime improvement strategy) in ascending order, and update the initial elite solution according to the sorting result to obtain the updated elite solution: 、 and .

[0093] judge and relationship, when When the current 、 and As the initial elite solution;

[0094] And reinitialize the worst solution periodically, that is, 、 and Reinitialize and combine with the current solution 、 and Take them as input and then execute the dual adaptive weight mechanism and rime search strategy in sequence until , end the evaluation and output the optimal threshold.

[0095] like Figure 2 As shown, the method of the present invention segments the image according to the optimal threshold value obtained by the improved rime algorithm, and maps each area of ​​the segmented image to different colors to obtain a color mapping image, so as to more intuitively display the segmentation result.

[0096] Example 2

[0097] This embodiment further limits the embodiment 1.

[0098] In this example, some corn germination seeds are selected to verify the effectiveness of the method of the present invention. The original corn seed germination images selected by the present invention are as follows: Figure 3 As shown, according to the method of the present invention, Figure 3 The optimal segmentation threshold of the image shown is 6. The image segmented according to the optimal segmentation threshold is as follows: Figure 4 shown.

[0099] like Figure 5 As shown, each area obtained by segmenting the image is mapped to different colors to more intuitively display the segmentation results.

[0100] according to Figure 3 、 4 From Figure 5 and Figure 6, it can be seen that the method of the present invention can effectively segment the germination status of corn, which is convenient for subsequent status detection and identification.

[0101] Example 3

[0102] This embodiment further limits the embodiment 1.

[0103] In order to comprehensively verify the optimization performance and generalization ability of the improved Rime algorithm (EARIME), this embodiment uses the Friedman two-way rank analysis of variance (Friedman test, abbreviated as FT) and the Wilcoxon signed rank test (Wilcoxon signed rank test, abbreviated as WSRT) to test the evaluation results of the improved Rime algorithm (EARIME) and the comparison algorithm in 30 test functions in the CEC2017 test function standard test suite, and ranks the algorithms involved in the comparison to intuitively display the average performance differences of different algorithms, further verify the effectiveness and robustness of the algorithm, and determine whether the improved algorithm is statistically significant.

[0104] First, this embodiment conducts comparative tests on the improved rime algorithm (EARIME) with the differential evolution algorithm DE, the particle swarm optimization algorithm PSO, the whale optimization algorithm WOA, the sine-cosine optimization algorithm SCA, the Harris Hawk optimization algorithm HHO, the Kepler optimization algorithm KOA and the mirage algorithm FATA. The comparative test results are shown in Table 1, where "15 / 3 / 12" in the fourth column of the second row indicates that in the test function evaluation results, EARIME performs significantly better than DE on 15 test functions (a total of 30 test functions), performs worse than DE on 3 test functions, and there is no significant difference in the performance between EARIME and DE on 12 test functions.

[0105] Table 1

[0106]

[0107] Secondly, this embodiment conducts comparative tests on the improved rime algorithm (EARIME), the sine-cosine algorithm and differential evolution hybrid optimization algorithm SCADE, the weighted differential evolution algorithm WDE, the improved grey wolf optimization algorithm IGWO based on differential evolution and elimination mechanism, the comprehensive learning particle swarm optimization algorithm CLPSO, the elite non-dominated sorting Harris Hawk optimization algorithm NSHHO, the dual adaptive enhanced whale algorithm RWDOA, and the chaotic mutation moth flame algorithm CLSGMFO. The results of the comparative tests are shown in Table 2:

[0108] Table 2

[0109]

[0110] Finally, this example conducts comparative tests on the improved rime algorithm (EARIME) and three RIME variant algorithms: the adaptive chaotic Gaussian rime optimization algorithm ACGRIME, the multi-objective rime optimization algorithm MORIME, and the dual-enhanced solution quality cross rime algorithm EECRIME. The results of the comparative tests are shown in Table 3:

[0111] Table 3

[0112]

[0113] The results in Tables 1, 2, and 3 show that EARIME achieves the best overall performance across all 30 test functions in the CEC2017 test function standard test suite. EARIME's average ranking in both the WSRT and FT tests ranks first among all algorithms, significantly outperforming the runner-up.

[0114] This example verifies that EARIME has a higher segmentation accuracy. Compared with the swarm intelligence optimization algorithms involved in the comparison, EARIME can retain more seed structure details and germination characteristics, and is significantly better than the comparison algorithms in multiple image evaluation indicators, proving that it can effectively handle the complex challenges of corn germ germination image segmentation. It also shows that EARIME has obvious advantages in overall optimization performance and stability.

Claims

1. A corn sprout image segmentation method based on an elite adaptive rime algorithm is characterized in that: The method comprises the following steps: S1. Obtain a corn sprout image dataset and perform preprocessing to obtain a grayscale image and a non-local mean image; S2, draw a two-dimensional histogram based on the grayscale image and the non-local mean image, and input it into the Kapur entropy function to obtain the objective function fobj; S3. Construct and improve the rime algorithm to obtain the improved rime algorithm: S31. Add an initialization elite solution module after the initialization module; The initialization elite solution module specifically comprises the following steps: sorting the initial fitness values ​​of the rime population in ascending order, and selecting the first three rime particles according to the sorting result as the initial elite solution: 、 and ; S32. Add a dual adaptive weight mechanism between the initialization elite solution module and the rime search strategy module; The dual adaptive weight mechanism is specifically: S321, calculate global search weight and local search weight : , ,in, Indicates the current evaluation number, Indicates the maximum number of evaluations; S322, according to Calculate the parameters used to control the search step size of the improved soft rime strategy , specifically: ,in, express A random number between Represents pi and the number of current function evaluations The product of represents the rounding function, Indicates soft rime parameters; S323, according to Calculate the trigger probability parameters of the soft rime improvement strategy respectively and Hard Rime Improved Strategy Triggering Probability Parameters ,in, , ,in, represents the square root function, Represents the normalized value of the fitness value of all rime particles, Indicates the The fitness value of each rime particle; S33. Using the elite reselection strategy to improve the soft rime strategy and the hard rime strategy, obtaining the soft rime improvement strategy and the hard rime improvement strategy; The soft rime improvement strategy is specifically as follows: ,in, Indicates the In terms of dimension, The optimal solution of the improved soft rime strategy for rime particles is searched. Indicates the In the dimension, the initial elite solution is selected according to the probability, where the selection 、 and The probability ratio is 6:3:1; The hard rime improvement strategy is specifically as follows: ,in, Indicates the In terms of dimension, The optimal solution of the hard rime improvement strategy for rime particles is searched; S34, update the elite solution after the rime search strategy module; S4, input the objective function fobj into the rime improved algorithm to obtain the optimal threshold; S5. Segment the grayscale image obtained in step S1 according to the optimal threshold value to obtain a segmented image.

2. The corn sprout image segmentation method based on the elite adaptive rime algorithm according to claim 1, characterized in that: The pre-processing specifically includes: sequentially performing grayscale processing and non-local mean filter processing.

3. The corn sprout image segmentation method based on the elite adaptive rime algorithm according to claim 1, characterized in that: The method of updating the elite solution after the rime search strategy module is as follows: sorting the fitness values ​​of the individuals in the population updated by the rime search strategy module in ascending order, and updating the initial elite solution according to the sorting result to obtain an updated elite solution: 、 and .

4. The corn sprout image segmentation method based on the elite adaptive rime algorithm according to claim 3 is characterized in that: The objective function fobj is input into the rime improved algorithm to obtain the optimal threshold, which is specifically: S41, initialize the rime population, , the upper limit of the search dimension , the lower limit of the search dimension and fobj; S42. Obtain the initial elite solution: 、 and ; S43, using the dual adaptive weight mechanism to calculate respectively: 、 and ; S44, when When implementing the soft rime improvement strategy, express Otherwise, execute step S45; S45, when When implementing the hard rime improvement strategy, express Otherwise, execute step S46; S46, Get the updated elite solution: 、 and ; S47, when When , the elite solution obtained in step S46 is used as the initial elite solution, and steps S43 to S47 are executed in a loop; Otherwise, the evaluation ends and the optimal threshold is output.

Images