Image Segmentation Based On Particle Swarm Optimization Extreme Learning Machine

The medical image segmentation has always been a difficult problem because the medical image has complex background,low noise ratio and uneven measurement.It is valuable to find a high accuracy and less time consumption method to segmenting medical images.As a random search algorithm,PSO boasts rapid convergence speed and strong random search capabilities,but is not suitable for managing the multidimensional mapping problem.In addition,the neural networks,as a complicated network structure,can deal with various large mathematical mapping problem,with good generalization,but has low algorithm efficiency with overfitting.To solve the problems above,the following contents are studied in this paper:1.Modified particle swarm optimization based on extreme disturbance(RPSO).The basic particle swarm optimization can work easily and efficiently,but with the defect of local optimum.In this paper,the reason for local optimum is analyzed,and the individual extremum and global extremum are searched by RPSO method according to change-stop steps(T0 and Tg)of individual extremum and global extremum of particle population during iterative process,which improves the random search ability and extends the search space.In addition,the complex multidimensional function experiment testifies the optimization ability with high efficiency.2.Modified particle swarm optimization algorithm for limit learning machine(RPSO-ELM).This paper introduces the basic idea of extreme learning machine algorithm(ELM),provides the principle of ELM training and classification,and then presents the RPSO-ELM algorithm by combination of RPSO and ELM.The RPSO-ELM algorithm uses input weight ?i and hidden layer paranoid bi to carry out the optimization by means of good multidimensional space random search capability of RPSO,obtaining the optimal ELM mode,which improves the generalization and classification performance of ELM.3.Practical application of RPSO-ELM algorithm in medical image segmentation.The sample set is created by pretreatment and feature extraction,and put into RPSO-ELM for the purpose of classification.After the classification is completed,the contour image is outlined by a series of morphological processing and conversion algorithm.Then the contour image is compared with standard results and basic ELM in terms of true positive rate,background correct recognition rate and background false recognition rate indicator,which indicates that the RPSO-ELM algorithm is feasible for practical medical image segmentation.