基于改进的SegFormer模型的遥感影像农田边界提取研究OA

Accurate and efficient automated extraction of farmland boundaries is a key tech-nology for the development of intelligent modern agriculture.To address problems such as large model parameter sizes and incomplete boundary extraction in existing methods,an im-proved SegFormer model is proposed by incorporating an attention mechanism and a feature pyramid module into the original SegFormer architecture.The proposed model was validated using the GID high-resolution remote sensing image dataset.The results show that the model achieved a mean Pixel Accuracy(mPA)of 95.22%and a mean Intersection over Union(mIoU)of 91.25%in farmland boundary extraction.Comparative experiments with DeepLabV3+,U-Net and PSPNet demonstrate that the improved SegFormer model out-performs these three benchmark models,with mPA improvements of 0.77%,3.11%and 5.72%and mIoU improvements of 0.90%,2.40%and 4.22%,respectively.Overall,the improved SegFormer model shows superior performance compared with representative deep learning methods such as DeepLabV3+,U-Net and PSPNet,indicating strong potential for practical application in farmland boundary segmentation.