Enhanced contextual representation with deep neural networks for land cover classification based on remote sensing images

Abstract

Classification tasks on land cover (LC) mapping are challenging due to the complex and heterogeneous characteristics of remote sensing images(RSIs). Current LC classifications are mainly based on deep convolutional neural networks (DCNNs), and previous works have been proven that spatial context can offer essential cues for performance improvement. However, they still have some drawbacks that limit context capture ability: the ambiguity of global context and lack of efficient context combination strategy. To address these issues, we develop a multilevel LC contextual (MLCC) framework that can adaptively integrate the effective global context with the local context for LC classification. The MLCC framework comprises two modules: a DCNN-based LC classification network (DLCN) and a multilevel context integration module (MCIM). By a well-defined deep network, DLCN could enhance the effective global context feature while weakening the ambiguous representation. Besides, MCIM enables adaptive combinate the global and local context under the guidance of uncertainty map in an efficient way. This collaboratively global–local contextual information further boosts the discriminate feature representation for effective and efficient LC classification. The experiments on LC datasets demonstrate that the proposed MLCC has superior capability in capturing contextual features and thus outperforms the existing methods.