Inspired by the phenomenon that fireflies can use light for information exchange, Xin-She Yang from Cambridge put forward a new type of swarm intelligence algorithm:firefly algorithm (FA). Firefly algorithm has many advantages including simple operation and self organization, but the firefly algorithm has relatively high time complexity and is sensitive to the initial distribution. Based on the detailed elaboration of FA algorithm, the main work of this paper is as follows:Through the optimization test on a typical multimodal function, the convergence and the final population distribution under different various parameters values of FA is analyzed. At the same time, through simulation experiment, the optimization performance of FA and other typical swarm intelligence algorithms is compared.On the issue of high time complexity, an improved algorithm based on the clustering strategy which is called Cluster-based Firefly Algorithm (CBFA) is put forward. The improved algorithm carries on a dynamic cluster to the population, and makes basic cluster as the basic unit of evolution, thus reducing the number of individual search times and the time complexity of the algorithm. At the same time, the improved algorithm puts forward differential evolution strategy for the general individuals in clusters, the centers of the clusters and the global optimal individuals respectively.For Firefly algorithm, the stability can be easily affected by the initial population distribution, so a new algorithm is proposed to improve the Firefly Algorithm-Opposition-based Learning Firefly Algorithm (OBLFA). The opposition-based learning is introduced in the population initialization to expand the search region and to improve the population distribution. This way, the accuracy and stability of FA is enhanced.In this work, two kinds of improved FA algorithm are applied to maximum entropy and maximum between-cluster variance based image threshold segmentation respectively, and the ideal effect is achieved. |