This study addressed the issues of low efficiency in traditional remote sensing interpretation, difficulties in acquiring high-precision regional data, and the challenge of nationwide products failing to meet the needs for fine-scale land classification analysis. Relying on the Google Earth Engine platform, this study integrated Landsat remote sensing images and multi-source classification products to construct a multi-temporal land use/land cover dataset for Hulun Buir City from 2014 to 2024. Based on this dataset, classification experiments were conducted using several deep learning models, including FCN, Deeplabv3+, U-Net, and VM-Unet. Additionally, the study further analyzed the spatial and temporal evolution characteristics of land use in the study area and the driving effects of natural and social factors on the change process. The results are as follows. (1) The constructed dataset effectively supports regional-scale classification tasks. The classification accuracy of the deep learning models reaches 87.44%, with a Kappa coefficient of 0.823, accurately reflecting the dynamic features of land use/land cover. (2) Over the past decade, the grassland area has shown a continuous decreasing trend, while the areas of arable land and impervious surface have significantly expanded. Forest land and water bodies have remained relatively stable overall. (3) Natural factors such as DEM and slope play a dominant role in the changes of arable land, forest land, and water bodies, while population growth and economic development have notably driven the expansion of impervious surfaces. Additionally, the comprehensive driving effects of multiple interacting factors are particularly evident.
土地利用/覆盖变化(land use and cover change,LUCC)是人类与自然互动的体现,不仅是全球环境演变的关键指标,也是衡量人类活动对自然环境影响的重要方式。这种变化反映了人类在不同阶段对土地的利用手段,并揭示了人与自然之间复杂的互动关系,为科学研究提供了丰富的数据支持[1⁃3]。
随着深度学习的发展,语义分割方法为土地利用/覆盖(land use and land cover,LULC)分类带来了新的突破。Shelhamer等[5]引入了全卷积神经网络(fully convolutional network,FCN)通过自动提取图像的高级特征,将图像分类任务转变为像素级的语义分割,大大提升了精度。Ronneberger等[6]提出U-Net模型,其作为FCN的改进版进一步优化了分割精度,尤其在边界处理上表现出色。袁盼丽等[7]通过使用Deep-Labv3+、U-Net等语义分割模型,对新疆莫索湾垦区的土地利用类型进行分类,深度学习算法在开展地物提取分类及动态变化监测工作时显示出良好性能。近年来,基于Transformer模型和Mamba模型的应用逐渐兴起,Transformer通过自注意力机制捕捉长程依赖关系,特别是在处理大范围区域和多尺度特征时,展现出了优异的性能[8⁃9];相比之下,Mamba作为一种在图像领域新兴的深度学习网络结构,利用结构化状态空间序列模型,在处理遥感影像的多尺度特征融合上具有明显 优势[10]。它能够有效地减少计算复杂度,同时提升土地利用/覆盖分类的精度和效率。尽管深度学习方法在遥感影像分类中取得了一定进展,但仍面临数据需求大、计算资源消耗高以及训练过程复杂等挑战。
ZHOUD C, ZHAOS Q, ZHUC. The Grain for Green Project induced land cover change in the Loess Plateau: A case study with Ansai County, Shanxi Province, China [J]. Ecological Indicators, 2012, 23: 88-94.
SHELHAMERE, LONGJ, DARRELLT. Fully convolutional networks for semantic segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(4):640-651.
[6]
RENNEBERGERO, FISCHERP, BROXT. U-Net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention.Cham: Springer International Publishing, 2015: 234-241.
KHANM, HANANA, KENZHEBAYM, et al. Transformer-based land use and land cover classification with explainability using satellite imagery[J]. Scientific Reports, 2024, 14(1): 16744.
[10]
CAOY C, LIUC C, WUZ H, et al. Remote sensing image segmentation using vision mamba and multi-scale multi-frequency feature fusion[J]. Remote Sensing, 2025, 17(8): 1390.
KATTENBORNT, EICHELJ, FASSNACHTF E. Convolutional neural networks enable efficient, accurate and fine-grained segmentation of plant species and communities from high-resolution UAV imagery [J]. Scientific Reports, 2019, 9(1): 17656.
[13]
LUX Y, ZHONGY F, ZHENGZ, et al. Multi-scale and multi-task deep learning framework for automatic road extraction[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(11): 9362-9377.
[14]
EGOROVA V, ROYD P, ZHANGH K, et al. Landsat 4, 5 and 7 (1982 to 2017) analysis ready data (ARD) observation coverage over the conterminous United States and implications for terrestrial monitoring [J]. Remote Sensing, 2019, 11(4): 447.
TARIQA, JIANGOY, LIQ T,et al.Modelling,mapping and monitoring of forest cover changes,using support vector machine,kernel logistic regression and naive Bayes tree models with optical remote sensing data[J].Heliyon,2023,9(2):e13212.
[21]
FOLEYJ A, DEFRIESR, ASNERG P, et al. Global consequences of land use[J].Science, 2005, 309(5734): 570-574.
VELDKAMPA, FRESCOL O. Reconstructing land use drivers and their spatial scale dependence for Costa Rica (1973 and 1984)[J]. Agricultural systems, 1997, 55(1): 19-43.