Developing a general post-classification framework for land-cover mapping improvement using high-spatial-resolution remote sensing imagery

Abstract

ABSTRACTIn this letter, a general post-classification framework (GPCF) is proposed to enhance initial results. Traditional post-classification techniques usually improve classification accuracy by considering the contextual information in a single classified image. In contrast to traditional techniques, the proposed GPCF aims to integrate multi-source classified images obtained through different classification approaches. In the proposed framework, the label of a central pixel is determined by its surrounding voting in each classified image. In this manner, the GPCF can integrate the advantages of different classification approaches. In our experiments, a hyperspectral image and an aerial image with high spatial resolution (HSR) are used to evaluate the proposed GPCF. Compared with two relevant post-classification approaches, the proposed framework can provide a land-cover map with lower noise in visual comparison and achieve higher classification accuracies. Therefore, the proposed GPCF presents better performance in HSR image classification.