Expansion and revival of coseismic landslides severely affect social reconstruction and economic recovery years to decades after a disastrous earthquake. Thus, automatic identification of plentiful coseismic landslides across a wide region is critical for post-disaster social and economic sustainability, and has attracted great attention from scientists and policy-makers worldwide. Intelligent recognition of coseismic landslides has always been great challenges due to small sizes, variable shapes, and diverse spectra of coseismic landslides and due to various complicated environmental backgrounds. In remote sensing images, coseismic landslides are difficult to discriminate from complex geological environments such as bare rock, unvegetated soil, cropland, and manual excavation areas. Therefore, it is in high demand to research advanced and precise recognition algorithms of coseismic landslides. This chapter proposes a novel semantic segmentation network of EGCN to realize relatively accurate landslide identification across a wide-area seismic zone. EGCN is capable of integrating both high-level and low-level context features as well as local spatial characteristics. It demonstrates remarkable adaptability to diverse geological environments and variable geometric features of coseismic landslides, thereby significantly enhancing the recognition accuracy. The meizoseismal region of the Ms 7.0 Jiuzhaigou earthquake is selected as an example to highlight the superiority of EGCN when applied in an extensive region.