Swin-UperNet: A Semantic Segmentation Model for Mangroves and Spartina alterniflora Loisel Based on UperNet

As an ecosystem in transition from land to sea, mangroves play a vital role wind and wave protection biodiversity maintenance. However, the invasion of Spartina alterniflora Loisel seriously damages mangrove wetland ecosystem. To protect scientifically dynamically, semantic segmentation model for Loise was proposed based on UperNet (Swin-UperNet). In Swin-UperNet model, data concatenation module make full use multispectral information remote sensing images, backbone network replaced with Swin transformer improve feature extraction capability, boundary optimization designed optimize rough results. Additionally, linear combination cross-entropy loss Lovasz-Softmax taken as function Swin-UperNet, which could address problem unbalanced sample distribution. Taking GF-1 GF-6 images experiment data, performance compared against that other models terms pixel accuracy (PA), mean intersection over union (mIoU), frames per second (FPS), including PSPNet, PSANet, DeepLabv3, DANet, FCN, OCRNet, DeepLabv3+. The results showed achieved best PA 98.87% mIoU 90.0%, efficiency higher than most models. conclusion, is efficient accurate synchronously, will provide scientific basis monitoring resource conservation management.