Road　extraction　from　remote　sensing　images　is　essential　for　autonomous　driving,　traffic　management,　and　map　updating.　Nevertheless,　the　result　of　road　extraction　might　become　fragmented　and　disconnected,　due　to　a　number　of　reasons　in　complex　environments　(such　as　high　vegetation　coverage　occlusions).　In　this　paper,　we　propose　a　road　extraction　method　based　on　asymmetrical　GAN　framework.　First,　the　generator　employs　an　asymmetric　encoder-decoder　structure　for　road　pixel　extraction.　In　order　to　limit　the　input　of　noisy　information,　the　simplified　decoder　eliminates　the　multi-level　cascade　operation　in　the　symmetric　structure.　Second,　the　discriminator　adopts　the　FCN-based　architecture.　In　order　to　ensure　that　the　extracted　road　information　is　more　connected,　we　add　topological　structural　supervision　to　the　discriminator　adopting　the　FCN-based　architecture.　Our　approach　performs　better　on　the　real-world　road　dataset　when　compared　to　a　number　of　advanced　models.　Final　results　demonstrate　that　compared　to　previous　methods,　our　method　significantly　enhances　Recall\F1\IoU.