An Elite Group Guided Artificial Bee Colony Algorithm And Its Application To Multi-threshold Image Segmentation

Artificial bee colony algorithm(ABC)is a popular global optimization algorithm in recent years.It simulates the foraging behavior of bee colony to achieve optimization.It has the characteristics of simple algorithm structure and excellent performance,and has attracted many researchers' attention and application.However,similar to other evolutionary algorithms,ABC also faces the disadvantages of low performance in solving complex optimization problems,such as slow convergence speed,low accuracy and long solution time.Therefore,this paper studies how to use the elite individuals in the population to improve the performance of the algorithm,proposes an elite group guided artificial bee colony algorithm,and accelerates it in GPU parallel.Finally,the algorithm is applied to solve the multiple thresholds gray-scale image segmentation problem.The main work is as follows:(1)In the classic ABC,individuals generate their offspring through the solution search strategy,but the performance of the algorithm is limited because of its strong exploration ability and weak exploitation capabilities.Therefore,based on the idea of elite individuals,this paper proposes an improved artificial bee colony algorithm(ENABC)based on the guidance of elite group,selects several better individuals from the population to form the elite group,and then designs two new search strategies based on the elite group,which are used in the stage of employing bees and onlooker bees,in order to balance the exploration and exploitation capabilities of the algorithm.Furthermore,based on the elite group,an improved neighborhood search operation is proposed,in which fine-grained search is carried out near the elite individuals in order to find better solutions and speed up the convergence of the algorithm.In order to verify the effectiveness of ENABC algorithm,it is compared with 8 excellent improved ABC algorithms on 50 test functions.The experimental results show that ENABC algorithm has better performance.(2)Compared with the traditional optimization algorithm,the running time of ABC for solving high-dimensional complex optimization problems is often unacceptable.Therefore,this paper studies a parallel ABC based on the GPU.The parallel version of the ENABC algorithm is realized by using the CUDA programming model of NVIDIA.By combining the characteristics of the ABC,a single thread is used to simulate a single individual,and the GPU is carried out for individual initialization operation,update operation,and calculate fitness function operation,thus reduces the running time of the algorithm.In order to verify the effectiveness of the parallel algorithm,we compared it with CPU serial version when CEC 2013 test function is 30,50 and 100 dimensions respectively.The experimental results show that GPU parallel version can accelerate up to 10.41,which effectively shortens the running time of the algorithm.(3)Multi-threshold segmentation of gray image is a difficult problem in image segmentation technology,and it is also an important part of image semantic meaning.For this reason,this paper proposes a multi-threshold segmentation method of gray image based on ABC,which uses Kapur entropy of image as the optimization objective function,encodes a group of segmentation thresholds into an individual,and then solves them by ENABC algorithm.Finally,a group of thresholds can make Kapur entropy maximum and complete image segmentation at the same time.In order to verify the validity of the segmentation method,experiments are carried out on BSDS 500 data set,and compared with exhaustive method and 3 different ABC algorithms.The experimental results show that the running time of ENABC algorithm is much shorter than exhaustive method when the segmentation accuracy is close to exhaustive method.Compared with other 3 ABC algorithms,ENABC algorithm has better segmentation accuracy and running time Performance.