基于优化Swin-UNet模型的高原山地牧场要素提取研究OA

The grasslands of the Tibetan Plateau are experiencing vegetation cover degradation under the combined effects of climate change and economic diversification,leading to a continuous decline in grassland carrying capacity.This necessitates the extraction and quantitative assessment of plateau mountain pasture elements using high-resolution remote sensing data.This study employs multi-temporal remote sensing images from GaoFen-2(GF-2)and GaoFen-7(GF-7)satellites to investigate grassland cover and habitat element extraction.Given that the original Swin-UNet model has limited capability in representing fragmented patches,complex terrain boundaries,and multi-scale grassland structures in plateau mountain pasture scenarios,which tends to result in local detail information loss and insufficient scale adaptability,this paper proposes an optimized Swin-UNet model based on the original framework for grassland classification and coverage estimation.Specifically,the neural network removes one hierarchical layer containing two Swin Transformer Blocks(STBs).It incorporates convolutional layers with same-dimension residual connections after each STB level to enhance local details.Meanwhile,an Atrous Spatial Pyramid Pooling(ASPP)module is integrated at the final skip connection to strengthen multi-scale contextual information fusion.Subsequently,spectral and spatial features extracted from imagery and the Digital Elevation Model(DEM)are incorporated into the optimized Swin-UNet model to classify grass-livestock balance conditions.Compared with the original Swin-UNet model's overall classification accuracy in the withered grass period(0.964)and the flourishing grass period(0.960),the improved model achieves 0.977 and 0.980 in the two periods,respectively,demonstrating significant improvement in overall classification performance.The model accurately reflects the distribution of mountain pasture elements across different seasons,enabling scientific assessment of grassland conditions and the grass-livestock balance in plateau mountain pastures.