Abstract
In recent years, high-resolution remote sensing semantic segmentation based on data fusion has gradually become a research focus in the field of land classification, which is an indispensable task of a smart city. However, the existing feature fusion methods with bottom-up structures can achieve limited fusion results. Alternatively, various auxiliary fusion modules significantly increase the complexity of the models and make the training process intolerably expensive. In this paper, we propose a new lightweight model called top-down pyramid fusion network (TdPFNet) including a multi-source feature extractor, a top-down pyramid fusion module and a decoder. It can deeply fuse features from different sources in a top-down structure using high-level semantic knowledge guiding the fusion of low-level texture information. Digital surface model (DSM) data and open street map (OSM) data are used as auxiliary inputs to the Potsdam dataset for the proposed model evaluation. Experimental results show that the network proposed in this paper not only notably improves the segmentation accuracy, but also reduces the complexity of the multi-source semantic segmentation model.
Highlights
The generation of high-resolution remote sensing images (RSI) has provided more convenient and detailed data sources for many civil applications, such as land classification, urban planning, and environmental monitoring
We demonstrate that fusing open street map (OSM) data with RSI can achieve better segmentation results than Digital surface model (DSM)
This may be due to the fact that DSM is officially released without modifications and more compatible with the original remote sensing data than the OSM that were edited by ourselves
Summary
The generation of high-resolution remote sensing images (RSI) has provided more convenient and detailed data sources for many civil applications, such as land classification, urban planning, and environmental monitoring. Most existing land classification methods are timeconsuming and expensive, and difficult to apply to fully explore the potential value of big remote sensing data. The classification methods using machine learning or deep learning tools have gradually become the mainstream approaches to high-resolution remote sensing semantic segmentation. Usually focus on the characterization of the image pixels and mathematical modeling of local features for clustering or segmentation. They require only regional spatial information and do not use high-level semantic information. A large number of excellent end-to-end single-source semantic segmentation models have sprung up. Networks such as SegNet [5], U-Net [6], PSPNet [7] and DeepLab series [8,9,10,11] have achieved excellent segmentation results with a single input source
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
