Landuse and land cover identification and disaggregating socio-economic data with convolutional neural network

Abstract

We demonstrated an innovative learning method of convolutional neural network (CNN) to identify landuse and land cover (LULC) patterns and extract features to disaggregate socio-economic factors by using remote sensing imageries at 30 m spatial resolution. The training labels were extracted from the historical LULC map to reduce the huge cost of labelling work, and to provide an inaccuracy but sufficient training dataset. The fully connected layer of the trained CNN was extracted as disaggregating features to map socio-economic factors of population and gross domestic product (GDP). Results indicate that current method can attain 92% overall agreement of LULC identification with the cross-validation of other products. The determination coefficient of disaggregating socio-economic factors can reach 0.945 for population density, and 0.876 for GDP density with the cross-validation at county level.

Keywords:

• Land use identification
• population mapping
• convolutional neural network
• Landsat 8

Acknowledgements

The source code can be found in https://github.com/yaojt/tree/master/land_use_identification.

Disclosure statement

No potential conflict of interests was reported by the author(s).

Funding

This research is funded by the Beijing Natural Science Foundation (Project No. 8151001), the National Natural Science Foundation of China (Project No. 41771561), and the China Scholarship Council.