Attribute weighted Naive Bayes for remote sensing image classification based on cuckoo search algorithm

Abstract

The Naive Bayes classifier(NB) is an effective and simple classification method for remote sensing image classification which is based on probability theory. However, in general, the contribution of each feature is different for classification and its attribute independence assumption is often invalid in the real world. The attribute weighted Naive Bayes(WNB) classifier might have better performance compared to NB, nevertheless, it is a hard and time-consuming work to learn the weight values for all features. Cuckoo search is a newly proposed meta-heuristic optimization algorithm which has been successfully applied for many parameter optimization problems. In the paper, a remote image classification approach is proposed, the attribute weight of which is learnt through cuckoo search algorithm (CSWNB in brief). In order to testify the performance of the proposed method, it is compared to some other evolutionary algorithms, such as attributed weighted Naive Bayes based on Genetic Algorithm (GAWNB), attributed weighted Naive Bayes based on Particle Swarm Optimization (PSOWNB) and attributed weighted Naive Bayes based on Water Wave Optimization (WWOWNB) etc. Experimental results demonstrate that the proposed approach has higher classification accuracy and more stable performance.