CN114819038A - Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator - Google Patents

Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator Download PDF

Info

Publication number
CN114819038A
CN114819038A CN202210389278.7A CN202210389278A CN114819038A CN 114819038 A CN114819038 A CN 114819038A CN 202210389278 A CN202210389278 A CN 202210389278A CN 114819038 A CN114819038 A CN 114819038A
Authority
CN
China
Prior art keywords
individual
individuals
algorithm
clustering
equal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210389278.7A
Other languages
Chinese (zh)
Inventor
刘昌云
段玉先
郭相科
李松
王刚
韦刚
张春梅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Air Force Engineering University of PLA
Original Assignee
Air Force Engineering University of PLA
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Air Force Engineering University of PLA filed Critical Air Force Engineering University of PLA
Priority to CN202210389278.7A priority Critical patent/CN114819038A/en
Publication of CN114819038A publication Critical patent/CN114819038A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/08—Computing arrangements based on specific mathematical models using chaos models or non-linear system models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Optimization (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Economics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Image Analysis (AREA)
  • Development Economics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Molecular Biology (AREA)
  • Educational Administration (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Nonlinear Science (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Biophysics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Biomedical Technology (AREA)

Abstract

The invention discloses a target clustering method based on Gaussian mapping and a mixed operator improved image cluster algorithm, which comprises the steps of (1) initializing a population by using a Gaussian chaotic sequence, and improving the diversity and the ergodicity of the initial population; (2) the individual positions in the population are updated by adopting a random walk operator and a mutation operator, so that the global exploration capability and the local development capability of the algorithm are improved; (3) comparing the fitness of each generation of individuals, preferentially reserving, and finding a solution to the actual problem; (4) all samples are divided into the clusters according to the distance relation between each sample and the clustering center, so that the convergence rate and the optimization precision of the target clustering are improved, and the method has a better practical effect and a beneficial effect on solving the problems of single peak and multiple peak.

Description

Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator
Technical Field
The invention relates to the field of information fusion, in particular to a target clustering method based on Gaussian mapping and mixed operator improved image clustering algorithm.
Background
In the JDL (joint directives of laboratories) model, target clustering belongs to the category of situation element awareness process in the JDL model, and the result thereof will directly affect the subsequent understanding and prediction of situation, and the target clustering needs to take space geometric factors, time factors, function attachment factors and the like into consideration, especially in the military field, the extraction of key elements should be completed first, and the factors in various aspects such as operational environment, human geography, enemy and my power distribution, operational strength and the like should be paid attention to, and the state information of the target is extracted and mined to provide input information for the target clustering. Secondly, the extracted target information is preprocessed, and the influence of different dimensions is eliminated. And then, according to information such as speed, distance, course, type and the like, dividing space groups of the targets in a certain space-time range. And then, dividing function levels according to different execution functions, dividing interaction groups according to tactical tasks, and finally distinguishing the attributes of the enemy and the my. Through target grouping, the military force structures, the organizational relationships and the combat systems of the enemy and the enemy can be abstractly described, the executed task conditions can be understood and researched, situation understanding, situation prediction and threat estimation can be conveniently carried out in the next step, meanwhile, key targets and key areas are focused, and the commander is assisted to form overall cognition on the battlefield situation.
In the prior art, the following is mainly adopted for target grouping: the method mainly comprises the steps that firstly, a similarity measurement algorithm is improved, the target grouping effect is improved, secondly, a plurality of algorithms are combined, the problems and the defects of a single algorithm are overcome, the functions are complemented, the structures are mutually matched, and the overall operation efficiency of the algorithm is improved.
The prior art has the following problems:
1) most battlefield space sample data are fuzzy, complex and huge, and information such as battlefield environment enemy position, speed, identity, responsible tasks and the like is often difficult to obtain, so that the battlefield space sample data have the capacity of 'not being known before being lost' under the condition that prior knowledge such as parameter threshold, cluster number, initial cluster center and the like is unknown, and the influence caused by uncertainty is effectively adapted. Most of the existing algorithms need to manually adjust parameters in advance, and the quality of parameter adjustment directly influences the final clustering effect, which has certain unpredictability and uncontrollable property, so that accurate clustering or clustering is difficult to achieve by adopting the existing algorithms.
2) In the wind and cloud changing battlefield environment, the fighting situation at each moment is changed. Therefore, it is necessary to take full consideration of time complexity, respond to a dynamically changing battlefield environment at a high speed, classify targets in a short time, reduce time cost while ensuring clustering accuracy, and overcome problems of high speed, low speed and high speed. The existing algorithm is low in convergence speed, easy to fall into local optimum, incapable of obtaining global optimum and influenced in classification effect.
3) In the battlefield environment of today, various camouflage and deceptive means are always concerned by all parties. In order to reduce the influence of bait bombs, false targets and the like on the situation evaluation process, an algorithm is required to effectively distinguish the distribution of outliers and noise points, and the method has good robustness and stability. However, the mode of generating the clustering center by the existing clustering algorithm is mostly influenced by the position of the initial position, and once the initial position deviates from the final actual position, the operation efficiency of the algorithm is reduced, and the final clustering effect is directly influenced.
Disclosure of Invention
Aiming at the existing problems, the invention provides a target clustering method for improving the image clustering algorithm based on high-speed mapping and mixed operators.
The technical scheme adopted by the invention is as follows:
the target clustering method based on the Gaussian mapping and the mixed operator to improve the image clustering algorithm comprises the following steps:
step S1: initializing parameters, specifically setting a data set U containing a target object, a population scale N, setting a cluster family number h, setting a clan number c in an image cluster algorithm, setting an attribute dimension D of the target object, and setting a maximum iteration number t max The initialization iteration time t is 0;
here, U denotes an input original data set, and U ═ x 1 ,x 2 ,...x q ...x N }，x q (1. ltoreq. q. ltoreq.N) represents a target object,
Figure BDA0003596142730000031
d represents the dimension of the properties of the target object,
Figure BDA0003596142730000032
the mth dimension information, which can be expressed as the qth target object, such as altitude, speed, heading, etc., information.
The way of calculating the upper and lower bounds of the population is as follows: after the original data set U is obtained, distinguishing each dimension, and sorting the values of the individuals in the population on the same dimension, wherein the maximum value is an upper bound, and the minimum value is a lower bound.
Step S2: generating an initialized image group by a Gaussian mapping method according to the set group size N and the attribute dimension D of the target object;
step S3: calculating and sequencing the individual fitness of the initialized elephant trunk generated in the step S2 to obtain the fitness value of the individuals in the initialized elephant trunk, recording the fitness value of the first c best individuals and the position of the best individual, respectively distributing the c best individuals to c clans, and dividing the elephant trunk into c clans;
the specific way of dividing elephant groups into c clans is to randomly allocate individuals to each clan after recording the fitness values of the previous c best individuals and the positions of the best individuals, and allocate c individuals as female individuals (best individuals) to each clan. Then, other individuals in all clans are adjusted and optimized along with the position of the female individual, and local search is needed after global exploration, so that the optimization convergence process is completed. And establishing clans according to subsequent steps and continuously adjusting the positions of individuals in the clans along with an iterative process, so as to finally achieve convergence and complete an optimization process.
Step S4; updating the individual positions by adopting an improved image group optimization algorithm;
step S5: calculating and sequencing the fitness of the individuals in the object group, and reserving the first h individuals with higher fitness values as elite individuals, wherein h is a set cluster group;
step S6: judging whether the loop times are reached or not, and whether the iteration times t are equal to the maximum iteration times t or not max ：
If t is less than t max If yes, t +1, and go to step S4 again;
if t is equal to t max Outputting and storing h elitism individuals in the step S5 elephant group;
step S7: and calculating the distance relation between the individuals in the image group and the h elite individuals to perform target grouping.
Further, the process of generating the initialization image group by the gaussian mapping algorithm in step S2 includes:
step S21: generating a chaos sequence eta (beta) through Gaussian mapping;
Figure BDA0003596142730000041
wherein eta (beta) and eta (beta +1) respectively represent the current solution and the future solution of the mapping sequence, eta (beta) belongs to (0,1), the initial solution is generated by pseudo-random numbers which are not 0, and then a Gaussian chaotic sequence is generated through iteration;
step S22: mapping the chaotic sequence generated in the step S21 to a solving space to obtain a population Z ═ η 1 ,η 2 ...... η i 1,2.. N), wherein the parameter N represents the population size;
step S23: then the individuals in the elephant trunk are initialized
Figure BDA0003596142730000042
Can be expressed as:
Figure BDA0003596142730000043
wherein the content of the first and second substances,
Figure BDA0003596142730000044
expressed as the value of the kth individual in the m-dimension, l m And u m Respectively are the minimum value and the maximum value on the mth dimension, k is more than or equal to 1 and less than or equal to N, N is the population scale, m is more than or equal to 1 and less than or equal to D, and D is the attribute dimension; the generated initialization image group X (0) { X } 1 ,x 2 ,...,x k ,...x N }。
Further, the step S4 of updating the individual positions by using the improved image group optimization algorithm includes:
step S41 comparing x ci,j And x best,ci Wherein
Figure BDA0003596142730000045
Is the fitness value-optimized individual in the clan ci, x ci,j Is the jth individual in ci clan.
Step S42, if x ci,j Is equal to x best,ci Then update the optimal individual x best,ci The position of the mobile phone is determined,
Figure BDA0003596142730000051
if x ci,j Is not equal to x best,ci Then update the individual x ci,j The position of the mobile phone is determined,
Figure BDA0003596142730000052
step S43: further updating worst individual x worst,ci ，
x worst,ci ＝x worst,ci +δ m r 1 +K (c)；
Wherein the content of the first and second substances,
Figure BDA0003596142730000053
and
Figure BDA0003596142730000054
respectively representing the current and latest position of the best individual in the ci clan of the elephant group in t iterations,
Figure BDA0003596142730000055
and
Figure BDA0003596142730000056
respectively represent individuals randomly drawn from the clan ci, and rand is [0,1 ]]Random number between, x worst,ci Represents the position of the individual with the worst fitness in the Ci of the clan, delta m Is the coefficient of variation, δ m ＝0.1*(u m -l m )，r 1 ,r 2 ,r 3 ,r 4 Are random numbers uniformly distributed from 0 to 1,
Figure BDA0003596142730000057
u 1 is [ -1,1 [ ]]Random variables of, t and t max Respectively representing the current and maximum number of iterations,
Figure BDA0003596142730000058
is the optimal solution at the t-th iteration, PSR is the set threshold, x Gbest Representing a global optimal solution, C (sigma) represents a random number conforming to Cauchy distribution, and the definition formula is as follows:
Figure BDA0003596142730000059
Figure BDA00035961427300000510
where a is a position parameter defining the position of the peak of the distribution, b is a scale parameter of half the width at half the maximum, and in a standard cauchy distribution, a is 0, b is 1, and F (σ; a, b) represents the cauchy distribution function.
Further, step S5 retains the individual processes with the higher fitness values:
Figure BDA00035961427300000511
wherein GBestX represents a globally optimal individual position,
Figure BDA00035961427300000512
representing the kth individual position produced in the t-th iteration.
Further, in step S7, dividing the target object into h clusters according to the distance relationship between the target object of the data set U and h elite individuals, which can be expressed as:
Figure BDA0003596142730000061
wherein x is n Representing the nth sample in the data set, N is more than or equal to 1 and less than or equal to N, N is the sample size, g i And the clustering center is a clustering center, wherein the clustering center is based on the top h elites obtained in the step S6, i is more than or equal to 1 and less than or equal to h, and h is the set clustering number.
After the initialization of the population in step S2 is completed, h individuals in the population are randomly selected as an initial clustering center, and each searched individual is a matrix consisting of h × D and represents a group of clustering centers consisting of h objects with D dimensions. Wherein D represents the attribute dimension of the target object, and h is the number of clusters to be input.
The calculation method of the fitness value is as follows: the sum of the distances (sum of squared error SSE) of the other individuals from the cluster center of the group is calculated, i.e.
Figure BDA0003596142730000062
C i Indicating the ith cluster. In the followingIn the iteration, the final goal is to get a set of cluster centers that minimize the sum of squared errors SSE, i.e.
Figure BDA0003596142730000063
And continuously optimizing each searching individual to converge to the global optimal solution to obtain the final clustering center.
Compared with the prior art, the invention has the following beneficial effects:
the invention aims to improve the convergence rate of target grouping and the stability of a system, and improve individuals with higher quality;
firstly, the improved image group optimization algorithm is adopted to group the targets, so that the convergence speed of the existing algorithm can be improved, the global optimization capability is realized, the problem that the existing algorithm is easy to fall into local optimization is avoided, and meanwhile, the image group optimization algorithm has the advantages of simplicity and high efficiency due to the fact that the control parameters of the image group optimization algorithm are few.
Secondly, aiming at the problems that the sample characteristics of the actual battlefield space are mostly fuzzy, complex and large in data volume, the Gaussian mapping method is adopted to optimize the initialized population, the initial clustering center is accurately and effectively estimated, and clustering under the condition without prior knowledge are obtained.
And thirdly, updating the individual positions of the population through a random walk operator and a mutation operator, and replacing a clan update operator and a separation operator in the original image group algorithm, so that the global exploration capacity of the algorithm is enhanced, the local optimal trap is better avoided, and the convergence rate is optimized.
Compared with the existing clustering method and the D-S evidence theory method, Bayesian inference, a fuzzy set method, a semantic-based method, an uncertain inference-based method, a support vector machine, a clustering method, an expert system and a template-based method, the convergence of the method is improved. Experiments on 9 data sets show that the convergence rate and the optimization precision of the method are superior to those of other 8 meta-heuristic algorithms, and the method has good stability. The Iris, Seeds and Aggregation data sets are visually displayed, and the clustering center precision obtained by the method is improved compared with other traditional dominant algorithms, so that the clustering effect is obviously improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of an improved image group optimization algorithm based on Gaussian mapping and mixture operators according to the invention;
FIG. 2 is a graph comparing the convergence curves of the improved image group optimization algorithm (GBEHO) based on Gaussian mapping and the blending operator and other algorithms under different data sets;
FIG. 3 is a boxed graph of improved quasigroup optimization algorithm (GBEHO) and other algorithms based on Gaussian maps and blending operators under different datasets;
FIG. 4 is a graph comparing results of an improved image group optimization algorithm (GBEHO) and a particle swarm optimization algorithm (PSO) based on Gaussian mapping and a mixture operator respectively executed on an Iris data set; (a) distributing the original data set; (b) a PSO algorithm cluster map; (c) a GBEHO algorithm cluster map;
FIG. 5 is a graph comparing results of performing improved image group optimization algorithm (GBEHO) and Genetic Algorithm (GA) based on Gaussian mapping and blending operators on the feeds data set, respectively; (a) distributing the original data set; (b) GA algorithm clustering graph; (c) a GBEHO algorithm cluster map;
FIG. 6 is a graph comparing results after running a Gaussian mapping and mixture operator based improved swarm optimization algorithm (GBEHO) and a tabu search based Greensis optimization algorithm (GWOTS) on an Aggregation data set, respectively; (a) distributing the original data set; (b) GWITS algorithm cluster map; (c) a GBEHO algorithm cluster map;
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
In the description of the present invention, it should be noted that the terms "first", "second", "third", etc. are used only for distinguishing the description, and are not to be construed as indicating or implying relative importance, and furthermore, the terms "horizontal", "vertical", etc. do not mean that the components are absolutely horizontal or overhanging, but may be slightly inclined. For example, "horizontal" merely means that the direction is more horizontal than "vertical" and does not mean that the structure must be perfectly horizontal, but may be slightly inclined.
In the description of the present invention, it should also be noted that, unless otherwise explicitly specified or limited, the terms "disposed," "mounted," "connected," and "connected" are to be construed broadly and may, for example, be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in a specific case to those of ordinary skill in the art.
Referring to FIGS. 1 to 6:
example 1, according to the scheme of figure 1:
the target clustering method based on the high-speed mapping and mixed operator improved image clustering algorithm comprises the following steps:
step S1: initializing parameters, specifically setting a data set U containing a target object, a population scale N, setting a cluster family number h, setting a clan number c in an image cluster algorithm, setting an attribute dimension D of the target object, and setting a maximum iteration number t max The initialization iteration time t is 0;
here, U ═ x 1 ,x 2 ,...x q ...x N }，x q (1. ltoreq. q. ltoreq.N) represents a target object,
Figure BDA0003596142730000091
d represents the dimension of the properties of the target object,
Figure BDA0003596142730000092
the mth dimension information, which can be expressed as the qth target object, such as altitude, speed, heading, etc., information.
The data sets used in the examples are Iris, Wine, Seeds, break, heart, CMC, Vowel, Two-moon, and Aggregation. Wherein Iris, Wine, Seeds, break, heart, CMC, Vowel are all from the University of California, European University of California, machine learning database of Irvine, UCI. Two-moon and Aggregation are Two artificial datasets. Specific information for the 9 data sets is shown in table 1:
TABLE 1
Figure BDA0003596142730000101
Maximum number of iterations t max Set to 200, the population size N to 10, and the number of clans c in the cluster-like algorithm.
Step S2: generating an initialized image group by a Gaussian mapping method according to the set group size N and the attribute dimension D of the target object;
after the initialization of the population in step S2 is completed, h individuals in the population are randomly selected as an initial clustering center, and each searched individual is a matrix consisting of h × D and represents a group of clustering centers consisting of h objects with D dimensions. Wherein D represents the attribute dimension of the target object, and h is the number of clusters to be input.
Step S3: calculating and sequencing the individual fitness of the initialized elephant trunk to obtain the individual fitness of the initialized elephant trunk, recording the fitness values of the first c best individuals and the positions of the best individuals, respectively distributing the c best individuals to c clans, and dividing the elephant trunk into the c clans;
the specific way of dividing the population into c clans is to randomly allocate individuals to each clan after recording the fitness values of the c best individuals and the positions of the best individuals, and allocate c individuals as female individuals (best individuals) to each clan. Then, other individuals in all clans are adjusted and optimized along with the position of the female individual, and local search is needed after global exploration, so that the optimization convergence process is completed. And establishing clans according to subsequent steps and continuously adjusting the positions of individuals in the clans along with an iterative process, so as to finally achieve convergence and complete an optimization process.
The calculation mode of the fitness value is as follows: the sum of the distances of the other individuals from the cluster center of the group (sum of squared errors SSE) is calculated as the fitness value of the searched individual, i.e.
Figure BDA0003596142730000111
C i Indicating the ith cluster. In subsequent iterations, the final goal is to get a set of cluster centers that minimize the sum of squared errors SSE, i.e.
Figure BDA0003596142730000112
And continuously optimizing each searching individual to converge to the global optimal solution to obtain the final clustering center.
Step S4; updating the individual positions by adopting an improved image group optimization algorithm;
step S5: calculating and sequencing the fitness of the individuals in the object group, and reserving the first h individuals with higher fitness values as elite individuals, wherein h is a set cluster group;
the sum of the distances of the other individuals from the cluster center of the group (sum of squared errors SSE) is calculated as the fitness value of the searched individual, i.e.
Figure BDA0003596142730000113
C i Indicating the ith cluster. In subsequent iterations, the final goal is to get a set of cluster centers that minimize the sum of squared errors SSE, i.e.
Figure BDA0003596142730000114
And continuously optimizing each searching individual to converge to the global optimal solution to obtain the final clustering center.
Step S6: judging whether the loop times are reached or not, and whether the iteration times t are equal to the maximum iteration times t or not max ：
If t is less than t max If yes, t +1, and step S4 is repeated;
if t is equal to t max Outputting and storing h elitism individuals in the step S5 elephant group;
step S7: and calculating the distance relation between the individuals in the image group and the h elite individuals to perform target grouping.
Preferably, the process of generating the initialization image group by the gaussian mapping algorithm in step S2 includes:
step S21: generating a chaos sequence eta (beta) through Gaussian mapping;
Figure BDA0003596142730000121
wherein eta (beta) and eta (beta +1) respectively represent the current solution and the future solution of the mapping sequence, eta (beta) belongs to (0,1), the initial solution is generated by pseudo-random numbers which are not 0, and then the Gaussian chaotic sequence is generated through iteration.
Step S22: mapping the chaotic sequence generated in step S21 to a solution space by the Gaussian mapping sequencer of size N generated in step S21Calculating to generate an initial image group population with the size of N to obtain a population Z ═ eta 1 ,η 2 ...... η i 1,2.. N), wherein the parameter N represents the population size;
step S23: then the individuals in the elephant trunk are initialized
Figure BDA0003596142730000122
Can be expressed as:
Figure BDA0003596142730000123
wherein the content of the first and second substances,
Figure BDA0003596142730000124
expressed as the value of the kth individual in the m-dimension, l m And u m Respectively are the minimum value and the maximum value on the mth dimension, k is more than or equal to 1 and less than or equal to N, N is the population scale, m is more than or equal to 1 and less than or equal to D, and D is the attribute dimension; the generated initialization image group X (0) { X } 1 ,x 2 ,...,x k ,...x N }。
Preferably, the step S4 of updating the individual positions by using the improved image cluster optimization algorithm includes:
step S41 comparing x ci,j And x best,ci Wherein
Figure BDA0003596142730000125
Is the fitness value-optimized individual in the clan ci, x ci,j Is the jth individual in ci clan.
Step S42, if x ci,j Is equal to x best,ci Then update the most significant individuals x best,ci The position of the mobile phone is determined,
Figure BDA0003596142730000126
if x ci,j Is not equal to x best,ci Then update the individual x ci,j The position of the mobile phone is determined,
Figure BDA0003596142730000131
step S43: further updating worst individual x worst,ci ，
x worst,ci ＝x worst,ci +δ m r 1 +K；
Wherein the content of the first and second substances,
Figure BDA0003596142730000132
and
Figure BDA0003596142730000133
respectively representing the current and latest position of the best individual in the ci clan of the elephant group in t iterations,
Figure BDA0003596142730000134
and
Figure BDA0003596142730000135
respectively represent individuals randomly drawn from the clan ci, and rand is [0,1 ]]Random number between, x worst,ci Represents the position of the individual with the worst fitness in the Ci of the clan, delta m Is the coefficient of variation, δ m ＝0.1*(u m -l m )，r 1 ,r 2 ,r 3 ,r 4 Are random numbers uniformly distributed from 0 to 1,
Figure BDA0003596142730000136
u 1 is [ -1,1 [ ]]Random variables of, t and t max Respectively representing the current and the maximum number of iterations,
Figure BDA0003596142730000137
is the optimal solution at the t-th iteration, and PSR is the set threshold, which is set to 0.2, x in this embodiment Gbest Representing a global optimal solution, C (sigma) represents a random number conforming to Cauchy distribution, and the definition formula is as follows:
Figure BDA0003596142730000138
Figure BDA0003596142730000139
where a is a position parameter defining the position of the peak of the distribution, b is a scale parameter of half the width at half the maximum, and in a standard cauchy distribution, a is 0, b is 1, and F (σ; a, b) represents the cauchy distribution function.
Preferably, step S5 retains the individual processes with the higher fitness values:
Figure BDA00035961427300001310
wherein GBestX represents a globally optimal individual position,
Figure BDA00035961427300001311
representing the kth individual position produced in the t-th iteration.
Preferably, in step S7, the target objects are divided into h clusters according to the distance relationship between the target objects of the data set U and h elite individuals, which can be expressed as:
Figure BDA00035961427300001312
wherein x is n Representing the nth sample in the data set, N is more than or equal to 1 and less than or equal to N, N is the sample size, g i And the clustering center is a clustering center, wherein the clustering center obtains h elites individuals 1 and h which are not less than h and are set as the clustering cluster number according to the ranking obtained in the step S6.
The beneficial effects are that, in the experiment, the improved quasical group optimization algorithm (GBEHO) based on Gaussian mapping and mixed operators is compared with other 8-element heuristic algorithms, which are respectively a particle swarm algorithm (PSO), a differential evolution algorithm (DE), a Genetic Algorithm (GA), a cuckoo search algorithm (CS), a Gravity Search Algorithm (GSA), a Bat Algorithm (BA), a ant lion optimization mixed K-means algorithm (QALO-K) based on quantum heuristics and a wolf optimization algorithm (GWOTS) of mixed tabu search, and the parameter settings of each algorithm are shown in Table 2:
TABLE 2
Figure BDA0003596142730000141
Figure BDA0003596142730000151
In addition to the above parameters, the maximum number of iterations t of all algorithms is determined max Set to 200, population size N is set to 10, and parameter c of GBEHO in table 2 is the number of clans in the swarm-like algorithm.
FIG. 2 is a graph comparing convergence curves of improved Bing-Cluster optimization algorithm (GBEHO) and Particle Swarm Optimization (PSO), differential evolution algorithm (DE), Genetic Algorithm (GA), cuckoo search algorithm (CS), Gravity Search Algorithm (GSA), Bat Algorithm (BA), Quantum heuristic-based ant-lion optimization hybrid K-means algorithm (QALO-K), and hybrid taboo search Grey wolf optimization algorithm (GWOTS) under Iris, Wine, Seeds, break, heart, CMC, Vowel, Two-moon, and Ageggrion datasets;
iteration is the act of repeating a set of procedures to obtain the best solution. When all the procedures of an algorithm are repeated once, this is called an iteration, and the result of each iteration provides an initial value for the next iteration.
The convergence curve may reflect the convergence rate and global search capability during the algorithm iteration. GBEHO stabilized on the Iris, Wine, Seeds, Breast, Heart and Aggregation datasets for the 20 th generation. Although GBEHO converges more slowly on the Vowel dataset, the quality of the solution found is higher. The results verify that GBEHO has relatively fast convergence speed and excellent global search capability. Compared with GBEHO, the performance of the meta-heuristic methods such as PSO, DE, GA, CS, GSA and BA is slightly worse. This shows that, because the generation mode of the initialized population is improved by adopting the gaussian mapping, the diversity and the ergodicity of the population are improved, and the generated initial value is closer to the global optimum. In addition, the individual positions of the population are updated by using the random walk operator and the mutation operator, and the clan update operator and the separation operator in the original image group algorithm are replaced, so that the convergence rate of the algorithm is obviously improved, the algorithm can jump out the local optimal solution, and the optimization precision is also improved.
FIG. 3 is a boxplot of improved constellation optimization algorithm (GBEHO) and other algorithms under different datasets based on Gaussian mapping and blending operators;
it can be seen that the box plot of GBEHO is the narrowest of all data sets. Clearly, GBEHO has a more stable clustering ability and improves population diversity by using a strategy of EHO and GBO blending. Furthermore, GBEHO produces the least outliers, which indicates that GBEHO is very robust. These facts indicate that the proposed algorithm can effectively circumvent local minima.
FIG. 4 is a graph comparing results of an improved image group optimization algorithm (GBEHO) and a particle swarm optimization algorithm (PSO) based on Gaussian mapping and a mixture operator respectively executed on an Iris data set; (a) distributing the original data set; (b) a PSO algorithm cluster map; (c) a GBEHO algorithm cluster map;
fig. 4 shows the visualization of clustering on the Iris dataset. GBEHO and PSO are the best two algorithms on the Iris dataset. Both algorithms can accurately divide the data set into three different clusters and can obtain relatively good solutions. In contrast, GBEHO finds a central point that is significantly closer to the real scene than PSO. This indicates that GBEHO has better performance. The center point of GBEHO finding was relatively stable at the 20 th generation in terms of number of iterations. This indicates that GBEHO has faster convergence speed and stability.
FIG. 5 is a graph comparing results of performing improved image group optimization algorithm (GBEHO) and Genetic Algorithm (GA) based on Gaussian mapping and blending operators on the feeds data set, respectively; (a) distributing the original data set; (b) GA algorithm clustering graph; (c) a GBEHO algorithm cluster map;
FIG. 5 compares the clustering results on the feeds dataset, where GBEHO and GA are the two dominant algorithms. It is clear that GBEHO achieves better center point location in generation 20 and the final result. In generation 20, GBEHO was able to extract the center point of the bottom left-most cluster, which GA was unable to do. It is clear that GBEHO can distinguish clusters more accurately than GA.
FIG. 6 is a graph comparing results after running the improved group-like optimization algorithm (GBEHO) based on Gaussian mapping and mixture operators and the Grouver optimization algorithm (GWITS) based on mixture tabu search on the Aggregation data set, respectively; (a) distributing the original data set; (b) GWITS algorithm cluster map; (c) a GBEHO algorithm cluster map;
the behavior on the Aggregation data set is shown in FIG. 6. For the two clusters above and to the left, GBEHO obtains a more accurate cluster center point. Both GWOTS and GBEHO can find the exact central point of the top right and bottom right two clusters. However, the partition of GBEHO in the lowest cluster is more pronounced. Although the black and magenta clusters in the figure are not accurately distinguished, this is due to the disadvantage of the conventional euclidean distance. From the viewpoint of overall convergence rate and clustering accuracy, GBEHO is relatively more advantageous.
It can be seen that the advantages in the above method are: (1) initializing a population by using a Gaussian chaotic sequence, and improving the diversity and the ergodicity of the initial population; (2) the individual positions in the population are updated by adopting a random walk operator and a mutation operator, so that the global exploration capability and the local development capability of the algorithm are improved; (3) comparing the fitness of each generation of individuals, preferentially reserving, and finding a solution to the actual problem; (4) all samples are divided into the clusters according to the distance relation between each sample and the clustering center, so that the convergence rate and the optimization precision of the target clustering are improved, and the method has a better practical effect and a beneficial effect on solving the problems of single peak and multiple peak.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the present invention in any way. Any simple modification, change and equivalent changes of the above embodiments according to the technical essence of the invention are still within the protection scope of the technical solution of the invention.

Claims (5)

1. The target clustering method based on the Gaussian mapping and the mixed operator to improve the image clustering algorithm comprises the following steps:
step S1: initializing parameters, specifically setting a data set U containing a target object, a population scale N, setting a cluster family number h, setting a clan number c in an image cluster algorithm, setting an attribute dimension D of the target object, and setting a maximum iteration number t max The initialization iteration time t is 0;
step S2: generating an initialized image group by a Gaussian mapping method according to the set group size N and the attribute dimension D of the target object;
step S3: calculating and sequencing the individual fitness of the initialized elephant trunk to obtain the individual fitness values of the initialized elephant trunk, recording the fitness values of the first c best individuals and the positions of the best individuals, and distributing the c best individuals to the c clans respectively;
step S4; updating the individual positions by adopting an improved image group optimization algorithm;
step S5: calculating and sequencing the fitness of the individuals in the object group, and reserving h individuals with higher fitness values as elite individuals, wherein h is the set cluster number;
step S6: judging whether the loop times are reached or not, and whether the iteration times t are equal to the maximum iteration times t or not max ：
If t is less than t max If yes, t +1, and go to step S4 again;
if t is equal to t max Outputting and storing h elitism individuals in the step S5 elephant group;
step S7: and calculating the distance relation between the individuals in the image group and the h elite individuals to perform target grouping.
2. The target clustering method based on gaussian mapping and blending operator improved image cluster algorithm of claim 1, wherein the gaussian mapping algorithm method in step S2 for generating the initialization image cluster comprises the following steps:
step S21: generating a chaos sequence eta (beta) through Gaussian mapping;
Figure FDA0003596142720000011
wherein eta (beta) and eta (beta +1) respectively represent the current solution and the future solution of the mapping sequence, eta (beta) belongs to (0,1), the initial solution is generated by pseudo-random numbers which are not 0, and then a Gaussian chaotic sequence is generated through iteration;
step S22: mapping the chaotic sequence generated in the step S21 to a solving space to obtain a population Z ═ η 1 ,η 2 ......η i 1,2.. N), wherein the parameter N represents the population size;
step S23: then the individuals in the elephant trunk are initialized
Figure FDA0003596142720000021
Can be expressed as:
Figure FDA0003596142720000022
wherein the content of the first and second substances,
Figure FDA0003596142720000023
expressed as the value of the kth individual in the mth dimension, l m And u m Respectively are the minimum value and the maximum value on the mth dimension, k is more than or equal to 1 and less than or equal to N, N is the population scale, m is more than or equal to 1 and less than or equal to D, and D is the attribute dimension; the generated initial population X (0) ═ { X } 1 ,x 2 ,...,x k ,...x N }。
3. The method for target clustering based on improved image group algorithm based on Gaussian mapping and mixture operator as claimed in claim 1, wherein the step S4 using improved image group optimization algorithm to update individual positions comprises:
step S41 comparing x ci,j And x best,ci Wherein x is best,ci Is the fitness value-optimized individual in the clan ci, x ci,j Is the jth individual in clan ci;
step S42, if x ci,j Is equal to x best,ci Then update the optimal individual x best,ci Position:
Figure FDA0003596142720000024
if x ci,j Is not equal to x best,ci Then update the individual x ci,j A location;
Figure FDA0003596142720000025
step S43: further updating worst individual x worst,ci ，
x worst,ci ＝x worst,ci +δ m r 1 +K；
Wherein the content of the first and second substances,
Figure FDA0003596142720000026
and
Figure FDA0003596142720000027
respectively representing the current and latest position of the best individual in the ci clan in the elephant trunk over t iterations,
Figure FDA0003596142720000028
and
Figure FDA0003596142720000029
respectively represent individuals randomly drawn from the clan ci, and rand is [0,1 ]]Random number between, x worst,ci Represents the position of the individual with the worst fitness in the Ci of the clan, delta m Is the coefficient of variation, δ m ＝0.1*(u m -l m )，r 1 ,r 2 ,r 3 ,r 4 Are random numbers uniformly distributed from 0 to 1,
Figure FDA0003596142720000031
u 1 is [ -1,1 [ ]]Random variables of, t and t max Respectively representing the current and maximum number of iterations,
Figure FDA0003596142720000032
is the optimal solution at the t-th iteration, PSR is the set threshold, x Gbest Represents a globally optimal solution, and C (σ) represents a random number that conforms to the Cauchy distribution.
4. The method for target clustering based on improved image clustering algorithm based on Gaussian mapping and mixture operators as claimed in claim 1, wherein step S5 retains the individual processes with higher fitness value:
Figure FDA0003596142720000033
wherein GBestX represents a globally optimal individual position,
Figure FDA0003596142720000034
representing the kth individual position produced in the t-th iteration.
5. The method for target clustering based on improved image clustering algorithm by Gaussian mapping and mixture operator as claimed in claim 1, wherein said step S7, according to the distance relationship between the target object of data set U and h elites, divides the target object into h cluster families, which can be expressed as:
Figure FDA0003596142720000035
wherein x is n Representing the nth sample in the data set, N is more than or equal to 1 and less than or equal to N, N is the sample size, g i And the clustering center is a clustering center, wherein i is more than or equal to 1 and less than or equal to h according to the top h of the sorted elite individuals obtained in the step S6, and h is the set clustering number.
CN202210389278.7A 2022-04-14 2022-04-14 Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator Pending CN114819038A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210389278.7A CN114819038A (en) 2022-04-14 2022-04-14 Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210389278.7A CN114819038A (en) 2022-04-14 2022-04-14 Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator

Publications (1)

Publication Number Publication Date
CN114819038A true CN114819038A (en) 2022-07-29

Family

ID=82537237

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210389278.7A Pending CN114819038A (en) 2022-04-14 2022-04-14 Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator

Country Status (1)

Country Link
CN (1) CN114819038A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116935133A (en) * 2023-07-31 2023-10-24 湖南中医药大学第一附属医院((中医临床研究所)) Cardiovascular disease classification method and system based on SPECT image recognition

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116935133A (en) * 2023-07-31 2023-10-24 湖南中医药大学第一附属医院((中医临床研究所)) Cardiovascular disease classification method and system based on SPECT image recognition
CN116935133B (en) * 2023-07-31 2024-04-16 湖南中医药大学第一附属医院((中医临床研究所)) Cardiovascular disease classification method and system based on SPECT image recognition

Similar Documents

Publication Publication Date Title
CN111814251B (en) Multi-target multi-modal particle swarm optimization method based on Bayesian adaptive resonance
CN111722643B (en) Unmanned aerial vehicle cluster dynamic task allocation method imitating wolf colony cooperative hunting mechanism
CN108549233B (en) Unmanned aerial vehicle air combat maneuver game method with intuitive fuzzy information
CN113741508B (en) Unmanned aerial vehicle task allocation method based on improved wolf pack algorithm
CN111256697B (en) Unmanned aerial vehicle flight path planning method aiming at path point clustering machine learning
CN110986960B (en) Unmanned aerial vehicle flight path planning method based on improved clustering algorithm
CN113220033B (en) Multi-heterogeneous unmanned aerial vehicle task allocation method based on improved multi-universe optimization algorithm
CN113435108B (en) Battlefield target grouping method based on improved whale optimization algorithm
CN111707273A (en) Rapid aircraft track planning and optimizing method based on genetic algorithm
CN114819038A (en) Target clustering method for improving image cluster algorithm based on Gaussian mapping and mixed operator
CN111121784A (en) Unmanned reconnaissance aircraft route planning method
Cui et al. Novel PIO algorithm with multiple selection strategies for many-objective optimization problems
CN116579443A (en) Personalized federal learning method oriented to data isomerism and storage medium
Sharma et al. Diversity preference-based many-objective particle swarm optimization using reference-lines-based framework
CN111797966B (en) Multi-machine collaborative global target distribution method based on improved flock algorithm
CN113625767A (en) Fixed-wing unmanned aerial vehicle cluster collaborative path planning method based on preferred pheromone gray wolf algorithm
CN114444389A (en) Air attack target dynamic threat assessment method based on combined empowerment and improved VIKOR
CN115034690B (en) Battlefield situation analysis method based on improved fuzzy C-means clustering
CN116934148A (en) Air attack target threat assessment method based on clustering combination weighting
CN113919425B (en) Autonomous aerial target allocation method and system
CN116400737A (en) Safety path planning system based on ant colony algorithm
CN116595864A (en) Multi-unmanned aerial vehicle task allocation method based on mean shift clustering algorithm optimization
Habiboghli et al. A solution to the N-queens problem using biogeography-based optimization
CN115661576A (en) Method for identifying airplane group intention under sample imbalance
CN112925317B (en) AUV path planning method based on improved brainstorming optimization algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination