Research On Image Analysis Based On Swarm Intelligence Optimization Algorithms

In recent years,swarm intelligence optimization algorithm has developed rapidly.Many excellent algorithms have appeared and achieved good results in practical applications.Image analysis is the basic technology of machine vision,including many technologies such as image enhancement,image fusion,image recognition and image retrieval.There are many demands in medicine,transportation,military and aerospace.Especially the development of intelligent robots,smart healthcare,smart city and many other industries and fields bring many challenges to image analysis.At present,image analysis based on swarm intelligence optimization algorithm has become an important research hotspot.Aiming at the large number of complex and varied image processing performance optimization problems,we select different characteristics of swarm intelligence optimization algorithm to solve the above problems.On the research of intelligence optimization algorithm,improve the operation mechanism of swarm intelligence optimization algorithm and the construction of fitness function,to improve the analysis of typical optimization problem in convergence and convergence precision in image;on the other hand,this paper tries to introduce other mechanisms,so that the algorithm has better ability to search performance and sample diversity,exhausted,mobile peak and other issues,and achieved satisfactory results.On the research of image analysis,according to the different characteristics of swarm intelligence optimization algorithm,we choose different optimization algorithms through theoretical analysis and experimental verification,to solve the problem of image enhancement,image fusion,image recognition and image retrieval process,and achieved good results.This paper attempts to systematically explore the better application of swarm intelligence optimization algorithm in the field of image analysis.The main research contents include the following five aspects.First,aiming at the problems of image enhancement,such as increasing contrast,denoising and improving image quality,we design an image enhancement and optimization model which integrates global information and local information,and use teaching and learning based optimization algorithm(TLBO algorithm for short)to optimize it.First of all,in order to solve the image enhancement method is easy to loss of detail,enhanced global local enhancement method is easy to add noise problems,this paper designs a kind of fitness function,can accurately reflect the quality of the image;at the same time,to deal with the complicated optimization problem,TLBO algorithm is easy to fall into local optimum and low convergence precision.In this paper,we improve the TLBO algorithm in the "teaching" stage and "learning" stage,so that we can better consider the diversity and convergence,thus improving the overall performance of the algorithm.The experimental simulation results show that the improved algorithm improves the quality of image enhancement.Second,particle filter face tracking algorithm has the problems of mobile peak optimization,particle degradation and diversity loss.This paper proposes a face tracking algorithm based on differential evolution particle filter.First,the niche technology is used to improve the differential evolution algorithm.Secondly,the algorithm is used to replace the resampling operation in the particle filter process.It can not only push particles to the high likelihood region,but also ensure the diversity of particles,so that the process is more consistent with the problem of non rigid and dynamic movement of face targets in face tracking.The experimental results show that the method has good tracking accuracy and stability.Third,when the objective function of NIN network image classification model is non convex,high-dimensional and nonlinear,we use gradient descent method to optimize the model,which is easy to fall into local optimum and diversity exhaustion.In this paper,a NIN model method,which is based on the gradient descent method,is combined with the particle swarm optimization(PSO)algorithm.In the first stage,we use gradient descent method to pre train the NIN network model,and get the solution as the location of the initial population of PSO algorithm in the fine-tuning stage.In the second stage,we use PSO algorithm to fine tune the model.The algorithm has the advantages of low adjustment parameters,fast convergence speed and high convergence precision,which can improve the convergence of the model.Experiments show that this method has good image classification performance.Fourth,in view of the requirement that image fusion is visual and easy to process,a multi focus image fusion method based on biogeography is proposed to optimize pulse distribution cortex(SCM)model.On the one hand,the regional integration mode based on two steps,the first step: through regional gradient calculation and comparison of values of theimage,we can clearly determine the area,complete the preliminary fusion;the second step:according to the fuzzy region by the spiking cortical model of regional differences in image fusion,this paper proposes a fusion rule,can effectively improve the fusion performance,obtain better fusion effect.On the other hand,for pulse firing cortical model,it is necessary to manually set parameters,which is not conducive to intelligent optimal solution.The biogeography optimization algorithm is used to optimize the SCM model parameters.The experimental simulation results show that the algorithm can effectively improve the quality of image fusion.Fifth,the relevance feedback image retrieval algorithm based on evolutionary algorithm has the problem of poor user preferences and too many parameter settings.In this paper,two image retrieval algorithms combined with teaching and learning optimization and relevance feedback are proposed.First,in order to improve the accuracy of image retrieval,combined with the nearest neighbor classifier,we improve the fitness function of teaching and learning optimization algorithm,enhance the local development ability,and propose the TLBO-RF-P algorithm for improving the precision.Second,in order to improve recall rate of image retrieval,we introduce the PSO algorithm with strong space exploration ability.We improve the TLBO teaching stage and stage respectively,improve the global exploration ability,and propose TLBO-RF-R algorithm to improve recall rate.The experiment shows that the two methods can better combine the user’s preference information to improve the performance of image retrieval.