A novel decision fusion method based on weights of evidence model

Abstract

With the increasing availability of satellite data from different sensors, the fusion of multi-sensor image data has been widely required in diverse applications. The development of new information fusion methods is one of the important research topics. In this work, we proposed a novel method of decision fusion based on weights of evidence (WofE) model for land-cover classification using multi-sensor data. The prior probability and conditional probabilities obtained from classification results of different data sources were fused to produce the logit posterior odd for each class by using WofE model. The final class label for a pixel was decided as the one with the maximum logit posterior odd for the pixel. The proposed method was evaluated in land-cover classifications using multi-sensor data from two examples. The results showed that the proposed method effectively combined multi-sensor data in land-cover classification and thus significantly outperformed classifications using single data and achieved classification result comparable with or better than the classification results obtained by existing decision fusion methods.

Keywords:

• decision fusion
• weights of evidence
• classification
• multi-sensor data