Cotton　is　an　economically　important　crop　that　plays　a　crucial　role　in　improving　human　standards　of　living.　Accurate　spatial　information　about　cotton　is　essential　for　efficient　cotton　production　and　management.　In　this　paper,　we　employed　a　robust　and　lightweight　model,　namely　SegFormer,　to　extract　cotton　planting　areas　from　high-resolution　remote　sensing　images.　SegFormer　combines　the　Transformer　with　a　Multilayer　Lightweight　Perceptron　(MLP),　which　not　only　exhibits　stronger　feature　representation,　but　also　has　a　smaller　network　size　than　the　traditional　convolutional　neural　networks　(CNNs).　We　conducted　experiments　to　extract　cotton　planting　areas　during　the　cotton　maturity　period　in　Shandong　and　Xinjiang　using　Google　Earth　images　with　a　resolution　of　1.19　meters.　To　demonstrate　the　effectiveness　and　accuracy　of　SegFormer,　we　compared　it　with　U-Net　and　Swin-Unet　networks.　The　results　showed　that　SegFormer　performs　well　in　both　areas,　and　the　extracted　cotton　field　boundaries　are　clear　and　smooth　with　complete　shapes.　Moreover,　the　overall　extraction　accuracy　exceeds　95%,　which　is　better　than　the　performance　of　U-Net　and　Swin-Unet.　We　concluded　that　the　semantic　segmentation　method　based　on　SegFormer　is　effective　and　robust　for　extracting　mature　cotton　planting　areas　from　high-resolution　remote　sensing　images.　　©　2023　IEEE.