关键词: 深度学习, 地物分类, 卷积神经网络, 遥感影像, 语义分割

Abstract: Land cover classification plays a crucial role in environmental monitoring, resource management,and urban planning. However,the classification process faces various challenges due to factors such as spectral similarity,noise interference,and the intermixing of natural and man-made objects. To improve classification accuracy and enhance the robustness of the model,this paper proposes a deep learning network based on the U-Net convolutional architecture,combined with the Transformer self-attention mechanism.Experiments conducted on the remote sensing dataset of Lanzhou city demonstrate that the proposed model outperforms PSPNet,DeeplabV3,Segformer,and Swin-T in terms of average classification accuracy (mAcc),mean intersection over union (mIoU),and mean F1 score (mF1). In addition to improving classification accuracy,the model achieves high inference speed,showcasing its potential for applications in complex land cover scenarios and offering new insights for remote sensing image classification.

Key words: deep learning, land cover classification, convolutional neural network, remote sensing imagery, semantic segmentation