Land Cover Classification by Using of Multi-source Remote Sensing Image Based on ELM

High-precision land cover classification can provide a reliable reference for urban and resource development. Remote sensing images are accessible, precise and provide wide coverage, and are thus widely used in land cover classification research. Based on Extreme Learning Machine(ELM), land cover classification was carried out for the Xinmiao bubble and its vicinity in Songyuan City, Jilin, China using single Sentinel-1A image, single GF-2 image and their combined image. The results show that Sentinel-1A image has high sensitivity to saline-alkali land in the study area. The combined image of Sentinel-1A and GF-2 had a better classification result when compared to a single remote sensing image, and the extraction ability of woodland and saline-alkali land was stronger. The overall classification accuracy was 84.27% , and the Kappa coefficient was 0.7865. Futhermore, the combined image of Sentinel-1A and GF-2 was classified using the Support Vector Machine(SVM) and Mahalanobis distance algorithm separately, the classification accuracies were lower than that of ELM. Land cover classification based on ELM for multi-source combined remote sensing image clearly improved classification accuracy providing a practical alternative to current processes.