A Modified Particle Swarm Optimization Algorithm And Its Applications In Image Clustering

Computational intelligence algorithms were inspired by the wisdom of nature.In the last decade, they have made a very wide range of applications. Particle swarmoptimization algorithm as a typical representative of intelligence algorithms hasattracted the majority of scholars’ attention, focusing on improving the performanceof PSO algorithm, and enhancing in different applications. Because of functionoptimization problems without professional restrictions, research results can beexchanged easily, so, it has become the first choice to verify the algorithmperformance; Image clustering problem is an important application problems, andmeet the requirement of information age, and it’s suitable for testing the performanceof different algorithms, then it can be applied to other fields, such as medical anddesign.In this paper, we analysis the existing PSO and propose improvement strategies,then apply it to function optimization problem and image clustering problem,specific work include the following.First, analysis and comparison several common particle swarm algorithm, andprovide ideas to improve algorithm.Analysis of the standard particle swarm optimization and improved PSOalgorithm process, principles and conclusions, including: proportionally randominitialization particle swarm algorithm, based on global center points particle swarmoptimization, the hybrid algorithm PSO combining with the differential evolution,dynamically changing inertia weight particle swarm optimization, the referencedistance of history particle swarm optimal algorithm, reconstruction inertia weightPSO, release and limit speed particle swarm optimization algorithm.Second, propose fitness guided PSO algorithm for solving functionoptimization problems.By the above algorithm, improved PSO algorithms are divided into: particleswarm diversity in the process and inertia weight dynamically changed. In theprocess of evolution, increasing the particle swarm diversity, avoid being earlymaturity, at the same time, regulating the global search capability and local searchcapability, and finally realizing convergence. This paper analyzes the candidates particles should be the nearest forexchanging information, and the fitness is higher than the one, particle swarmalgorithm is proposed based on the fitness guide. Without increasing the parametersand human intervention, base on the global optimal solution, the historical bestsolution, adjacent particles, change the position of the particle in order to increasethe particle diversity. From constraint function optimization problems andno-constraint functions problems, which can be proved those algorithm isadvantages or disadvantages.Third, improve fuzzy sets particle swarm algorithm for solving image clusteringproblem.This paper analyzes base on fuzzy sets particle swarm algorithm for solvingimage clustering problem, then improve particle swarm algorithm: principalcomponent analysis to reduce the underlying feature for feature selection process;the center of the image using for initialization vector clustering process; fitness guidePSO updating particle swarm. Through two groups experiments, which can beproved the algorithm is feasible.