Integrating bottom-up classification and top-down feedback for improving urban land-cover and functional-zone mapping

As two kinds of basic units of cities, land-cover objects and functional zones play different but totally important roles in urban mapping and studies. Recent several years have witnessed significant improvement in their classification methods, e.g. geographic object based image analysis (GEOBIA). However, these methods focus mainly on bottom-up classifications from visual features to semantic categories but they ignore top-down feedbacks which are capable of optimizing classification results. To resolve the issue, this study presents an iterative method which integrates bottom-up and top-down processes for land-cover and functional-zone classifications. First, hierarchical semantic cognition (HSC) is employed to make bottom-up classification for land covers and functional zones. The HSC is essentially a hierarchical Bayesian model which links visual features, land covers, spatial object patterns, and functional zones together with a hierarchical structure. Then, a top-down feedback method, inverse hierarchical semantic cognition (IHSC), is proposed to optimize the initial classification results. Finally, the two processes are carried out iteratively to generate more and more accurate results. To verify the effectiveness of this method, we conducted it in Beijing, China. Experimental results indicate that the method produces accurate classification results of land covers and functional zones, and improves their accuracies by 9.9% and 6.5% respectively. Accordingly, our method combines bottom-up classification and top-down feedback and can significantly improve land-cover and functional-zone mapping results, thus can be regarded as a novel paradigm of urban mapping.