The Verification of Land Cover Datasets with the Geo-Tagged Natural Scene Images

Land cover is important for global change studies, and its accuracy reliability are usually verified by field sampling, which costs a lot. A method was proposed the verification of land datasets with geo-tagged natural scene images using convolutional neural network. The nature were firstly collected from Use Cover Area frame Survey (LUCAS) crowdsourcing platform Flickr, then classified according to Classification System. Nature Scene Image (NSIC) model based on GoogLeNet Inception network recognition constructed. Finally, in UK, as area, European Space Agency Climate Change Initiative (ESA CCI-LC) Global land-cover product fine classification system (GLC-FCS) NSIC-Inception image set. results showed that overall LUCAS very close product, 94.41% CCI LC 92.89% GLC-FCS, demonstrating feasibility NSIC model. In addition, VGG16 ResNet50 compared Inception. differences between Flickr discussed regarding image’s quantity, spatial distribution, representativeness, so on. uncertainties arising resolution different explored GCL-FCS. application has great potential support improve efficiency verification.