Skip to Main Content
An attempt has been made in the paper to find globally optimal cluster centers for remote-sensed images with the proposed Rapid Genetic k-Means algorithm. The idea is to avoid the expensive crossover or fitness to produce valid clusters in pure GA and to improve the convergence time. The drawback of using pure GA in the problem is the usage of an expensive crossover or fitness to produce valid clusters (Non-empty clusters). To circumvent the disadvantage of GA, hybridization of GA with k-Means as Genetic k-Means (GKA) is already proposed[GKA, Fast,Flash]. The Genetic k-Means Algorithm always finds the globally optimal cluster centers but the drawback is the usage of an expensive fitness function which involves σ truncation. The Rapid GKA alleviates the problem by using a simple fitness function with an incremental factor. A k-Means operator, one-step of k-Means algorithm, used in GKA as a search operator is adopted in this paper. In Rapid GKA the mutation involves less computation than the mutation involved in GKA and Fast GKA(FGKA). In order to avoid the invalid clusters formed during the iterations the empty clusters are converted into singleton cluster by adding a randomly selected data item until none of the cluster is empty. The results show that the proposed algorithm converges to the global optimum in fewer numbers of generations than conventional GA and also found to consume less computational complexity than GKA and FGKA. It proves to be an effective clustering algorithm for remote sensed images.