Fully convolutional networks for the classification of historical panchromatic aerial photographs

Open Access

Abstract

Remote sensing images provide the capability to obtain information of various landcover classes. This information is useful to the process of urban planning, socio-economicmodelling and population studies. There is a need for accurate and efficient methods to obtainthis information, particularly in regions where there is a scarcity of reference data. In thiswork, we develop a methodology based on fully convolutional networks (FCN) that is trainedin an end-to-end fashion using aerial RGB images only as input. The experiments are conductedon the city of Goma in the Democratic Republic of Congo. We compare the results to a state-of-the art approach based on a semi-automatic Geographic object image-based analysis(GEOBIA) processing chain. State-of-the-art classification accuracies are obtained by bothmethods whereby FCN and the best baseline method have an overall accuracy of 91.24% and89.34% respectively. The maps have good visual quality and the use of a FCN skip architectureminimizes the rounded edges that is characteristic of FCN maps. Finally, additionalexperiments are done to refine FCN classified maps using segments obtained from GEOBIA.This resulted in improved edge delineation in the FCN maps, and future work will involveexplicitly incorporating boundary information from the GEOBIA segmentation into the FCNpipeline in an end-to-end fashion. ; info:eu-repo/semantics/published