Kinetic-molecular Theory Optimization Algorithm And Its Application In Image Segmentation

The optimization problem is a research focus in scientific research and engineering applications.Looking for an optimization technology,which is rapid,stable and effective,is a topic that has long been discussed by all walks of life.Intelligent optimization algorithm,due to simple,fast computing,good optimizing effect,has attracted widespread attention of scholars at home and abroad.Some mechanisms in the kinetic molecular theory,such as attraction,repulsion and fluctuation,make it possible that optimization algorithms can consider both the convergence and the diversity of the population.So this paper proposed a new optimization algorithm based on the kinetic molecular theory,and focused on its application in image segmentation.Most of the existing intelligent optimization algorithms are poor in diversity and easily fall into local minima.Inspired by the kinetic molecular theory,this paper proposed a new optimization algorithm(kinetic-molecular theory optimization algorithm).The algorithm has designed attraction operator,repulsion operator and wave operator.Simulate molecular attraction,so that the particles can converge to the optimal value.Simulate molecular repulsion,so that the algorithm can maintain the diversity of the population.Simulate random thermal motion of molecules,so that the algorithm can always have global searching ability.Performance test shows that,this algorithm has obvious advantages in the aspects of solution quality,robustness,population diversity,convergence speed and so on.Due to the lack of local searching mechanism,the solution accuracy of kinetic-molecular theory optimization algorithm needs to be improved.And the misguidance may occur when the current optimal values is a local extremum.In view of the important role of elites in the optimization process,we improved the algorthm based on co-evolution and elite strategy,and then proposed a M elite synergy kinetic-molecular theory optimization algorithm.The algorithm uses M elites to avoid the misguidance,improves the convergence precision by learning and collaboration between the elites,and uses a new wave operator to prevent the algorithm falling into premature by dimension.Simulation results show that the improved algorithm has good performances in terms of solution accuracy,stability,ability to solve high-dimensional function and so on.Because of the exhaustive search on the threshold space,traditional multi-threshold segmentation algorithms have low computational efficiency.In view of the excellent performance of kinetic-molecular theory optimization algorithm,we tried to use the algorithm to solve multi-threshold problems.Based on Kapur's entropy and Otsu's criterias,we preliminary studied how to use use the algorithm to solve multi-threshold problems.Compared with the mainstream intelligent optimization algorithms,such as BF and DE,the algorithm is rapid,stable,and can segment the images well.Due to complex conditions in the industrial production of alumina,rotary kiln flame image often contain noise.If we used the above method to segment these images,then anti-noise capacity of the method appears to be poor.To overcome this problem,we improved Otsu method based on histogram oblique segmentation and proposed an improved Otsu method for segmentation of rotary kiln flame image.To reduce the amount of calculation and be convenient for multithreshold extension,a simplified distance measure function was used as the threshold selection criterion.To further enhance the anti-noise capability,post-processing based segmentation method was adopted here.And kinetic-molecular theory optimization algorithm was used to improve the calculating speed.The segmentation tests of alumina rotary kiln flame image from a factory verified the validity of the algorithm.Although the improved Otsu method has certain anti-noise capacity,its performance is still to be improved.So a new Otsu method based on projection of cross section was proposed in this paper.This method selects the optimal threshold value based a new histogram,and then applied a post-processing strategy based on threshold to process the segmented image.In order to get the regions of white matter,gray matter,cerebrospinal fluid from MR brain images,we extended it to multi-threshold and applied the M elite synergy kinetic-molecular theory optimization algorithm to select the optimal thresholds.The experimental result shows that this method has significant improvements in computational efficiency and anti-noise capability,and can segments the images corrupted with different noise well.This paper proposed a kinetic-molecular theory optimization algorithm,analyzed its strengths and weaknesses,improved it based on M elite strategy,and then applied it to image segmentation of the rotary kiln flame image and MR brain image.